During the course of this meeting, we may make projections or other forward-looking statements regarding future events or the future financial performance of the company and the industry. We wish to caution you that such statements are predictions and that actual events or results may differ materially. We refer you to the documents the company files on a consolidated basis from time to time with the Securities and Exchange Commission, specifically the company's most recent Form 10-K and Form 10-Q. These documents contain and identify important factors that could cause the actual results for the company on a consolidated basis to differ materially from those contained in our projections or forward-looking statements. These certain factors can be found in the investor relations section of Micron's website. Although we believe that the expectations reflected in the forward-looking statements are reasonable, we cannot guarantee future results, levels of activity, performance, or achievements.
We are under no duty to update any of the forward-looking statements after the date of the presentation to conform these statements to actual results.
Good morning, welcome to our 2015 Winter Analyst Conference. For those of you in the room, thank you very much for making the trek to balmy Arizona. I'm imagining for some of our northeastern friends in the room, it wasn't that hard a trip actually arriving here, given the weather conditions we keep reading about. For those of you in the webcast, again, thank you for joining us. We are very excited to have you here today because we think that the message that we're delivering around Micron's future is really one of excitement and opportunity. When you think about where we're coming from, we feel the company is positioned probably better than ever in the history of Micron.
What we want to talk to you today about why we think that and the investments we're making, investments in technology, the investments in market enablement and product development, the investments in future solutions, and how memory will play in the architecture of future technology, and also our ability financially to compete like we've never had before. Today's agenda is set up very much like that. We want to talk about the positions, technology, and our financial strength to invest. After my presentation this morning, Scott DeBoer, our Vice President of Research and Development, will come up and talk about the technology roadmap at Micron, both in the short term and long term.
Each of our BU leadership will talk about their individual markets opportunities, both in terms of enablement, attractiveness of each of the investments we're making in their sectors, as well as the product development efforts we're driving with our customers. After the BUs speak, Brian Shirley, who's our Vice President of Memory Solutions, is going to come up and talk a little bit longer term about where memory is going to play and advanced memory technology will play in future systems, end systems, future end segments, Internet of Things, strategic data, cloud analytics, emerging storage applications. That segment will be a little bit more forward-looking. Finally, of course, Mark Durcan is going to come up and talk about the health of the company financially and how we see the markets and the opportunity for Micron going forward. How we're positioned.
Coming off of a record year in 2014, where we achieved over $16 billion in revenue, had our most profitable year. You can kind of look back and see how we arrived there. Micron has played the role of innovator, bringing technology to market, of consolidator in the industry, as you can see over the timeline on this chart, as well as Micron has played the role of engaging with customers and bringing solutions to market. When we talk about a shift in our business, driving to more solutions orientation, we've had a history of doing that in DRAM and moving now into NAND, and our total business is well-positioned for this going-forward model we think is more attractive than ever in the memory sector. We also continue to see the memory market condition is very favorable.
We talk a lot about consolidated supply base, the competitors in the industry. I think what's important about that isn't just in terms of capacity, but it's also the scale, the ability to invest long term to take advantage of these opportunities. The scale we're at $16 billion in revenue, the ability to invest at the right level to compete is really something that's enabling Micron to take advantage of the market. As you might imagine, a few years back, at a dramatically lower revenue base, our ability to invest with the same portfolio of others was somewhat challenged. Today, we've never been in a stronger position given the market consolidation and how we are positioned in that base.
On the supply growth area, I'll talk to this in another slide, my follow-on slide here, but suffice to say that we think it's just more rational in terms of the investments being made in supply. We think this will maintain going forward for a variety of reasons. Let me just hold that for my next slide here. I think the most attractive part about the business for us on this, of the industry structure, is the continued development of these diversified end market segments. No longer is the memory business limited by the behavior of any one market segment. As a matter of fact, I think that our ability to navigate segments and optimize our business for each of the segment opportunities, both from a manufacturing and go-to-market perspective, are one of the key criteria for our success going forward.
Again, we continue to see the memory business and the market conditions as very favorable for us, and we look forward to investing as such. I said that I'd get back to the supply growth piece of the memory industry. In DRAM today, you can see today that the bit growth in DRAM is well towards the lower end of historical lows, and we think that will continue. Why do we think that will continue? First and foremost, the complexity of newer process migrations. It's kind of interesting that the actual benefits, the cost benefits, while still financially attractive at some level, are not nearly what they once were. Yet, the complexity of doing so is higher. When we look at DRAM in the market today, we don't see any massive change in the supply structure.
We think that we will continue to see a pretty balanced market in DRAM, which will allow for good returns. On the NAND side, it's still relatively low historically, obviously. This includes what we know about publicly about people's capacity plans. We still think this will be measured against financial returns because the end market applications are attractive enough for us to generate strong returns. I think what's different about DRAM and NAND is that in NAND, the demand picture is much more elastic. I think you've heard us say on calls and prior meetings that the elasticity in NAND is higher, and you may see more volatility. We also think the end markets in NAND, whether it be storage or mobile, our continued growth in consumer, we think these end markets will continue to grow in terms of the appetite for NAND.
It does have a lot to do with the migrations of technology and the available supply. We think NAND, in the near term, is relatively constrained. Constrained because of the trade-off between the economics inside the factories of 3D conversions, and also because of the current landscape and the people who have access to the technology. We see NAND as constrained over the near to long term in the business, and when you have a period like this, we think that it'll continue to remain healthy for some time. Interestingly enough, on the DRAM side, we continue to see strong demand, probably more so than we might have thought two years ago in terms of some interesting end markets. Just like in mobile phones, for example.
I don't think any of us thought two years ago that we were going to sit here and see mobile phone configurations of 3 gigabytes per unit, per phone. When we think about end markets and servers, when you talk to server manufacturers and they can't get enough bit growth, and they can't put enough DRAM in servers for optimizing performance. We're very pleased on the DRAM side for the demand picture of our end markets. Again, very excited about long-term growth and upside in the NAND solutions side of our business. It's interesting, we talk about solid-state drives a lot, both in the client and the enterprise, but the penetration of that today, given the cost trade-offs, is still relatively low to what the opportunity is.
We feel, again, very favorable markets in terms of supply and demand equation for the memory business as we see it today and going forward. What does 2015 bring for us? Well, in one word, it's about execution. First and foremost, on the technology front, we have numerous process and technology capability milestones we're trying to hit as a company, and we're excited about that. Without stealing much of Scott's thunder on the technology area, we feel we're in a fantastic position, both on the DRAM side and within the landscape of upgrading our factories and innovating with new technology, as well as in NAND, both at the product level and the technology level. Product level being the enablement of TLC, and at the technology level with vertical NAND. Again, Scott will update you here shortly on that.
In the technology area for us, it's a big year on execution, on bringing these products to market and getting them into manufacturing and out to our customers. In terms of product enablement, taking this core memory and advancing our solutions capability is a key priority for our company. Not only in R&D, but within the BUs and the engineering teams, and the adjacent packaging organizations, controller teams. Our company is focused on how to add value to our core technology. It's not just because that sounds great and it represents higher-margin opportunities. It's because the market is asking for that. Our customer relationships are shifting.
They're shifting in many ways, but perhaps the most significant way is as you think about those diversified markets I mentioned earlier, the discussions we're having with our customers are about how do we enable you, Mr. Customer, to develop an advantage with Micron Technology? No longer is that conversation about, "Hey, I woke up this morning and I saw the DRAM price exchange had a price of this. Let's talk about what our price should be." It's about how can your memory technology give me a differentiation when I'm out competing in the marketplace? That discussion requires us to think about how do we enable those customers of ours to drive differentiation. Again, it's core memory and then product enablement capabilities that we're investing in as a company. Finally, operational flexibility.
What's unique today in the memory business is that we need to have this flexibility because no one segment for us dictates or dominates our financial performance or how we go to market. We need to be able to react to demand and supply indications at an aggregate level of all these businesses. When we think about our capabilities at an operations level, whether it be manufacturing or assembly and test or supply chain, we have to be reacting to market demand signals to shift our capabilities to deliver where that demand is. The flexibility to do so has never been more critical.
The benefit of that is the diversification gives us very stable business platform going forward. We continue to invest and be able to be flexible and adapt at meeting our customers' needs across a whole variety of segments, which is attractive for us from a returns perspective. What are we investing in? We're making sizable investments in this capability. When you think about Micron and what we're looking at from an R&D perspective or a capital perspective, it's really the capability to take this core technology to do what I said, to add value, to develop solutions that are unique for our customer that only Micron can deliver. That really comes into 3 categories as we think about these investments. First and foremost, emerging memory. What you're going to hear Scott talk about in emerging memory is that memories will be developed for specific application areas.
I think about storage and the role of storage in performance and power, reliability, what customers are really looking for, and the ability to develop core technology that fits to a specific application is a much different climate than you might have seen in the memory business in the past. Then what do we do with that memory? How do we integrate that memory? Whether it be packaging or controllers or firmware, we're investing very heavily and adding resources in the company and capabilities to take that memory and again, do something unique with it around a subsystem or system-level capability to drive enhanced differentiation and greater value. Finally, you might imagine that years ago, delivering quality to a PC company was our top priority. It was.
Now, as you think about the flexibility I commented on earlier and the importance of quality at a much higher level, when you think about automotive companies or enterprise storage or networking companies, investing in those capabilities to deliver not only the best technology in the memory industry, but world-class solutions at the highest level of quality. That flexibility and that performance and that reliability will allow us to be a world leader in differentiated memory solutions. On the right-hand side of this chart, you also see an investment in people. As our business has changed, we've continually looked for ways to strengthen our leadership team. I had a number of conversations last night because people have noted, wow, there's a lot of new, unfamiliar faces here at the dinner from Micron.
We will continue to invest as we bring in what we think are the best world-class executives to help us achieve our goals of shifting our business to differentiated value-add capabilities, whether it be in storage or advanced computing architecture, whether it be in assembly, manufacturing, supply chain capabilities. Areas that we need to upgrade and enhance and build capabilities on a great team to lead us to this new business opportunity we think will make us the world's best memory company. We're very excited at Micron to have you here today and look forward to a great morning. With that, I'll hand it over to Scott DeBoer. We'll be back with Q&A from Mark, with Mark and I later. Thank you.
Morning. Thanks, Mark. I'm going to start off today just with a high-level view of a lot of the different technologies, and then I'll go into several of these in more detail and try to touch on some of the big themes for us. As Mark mentioned, this upcoming year is a huge year, an exciting year for us in terms of technology execution, and we think we're really well-positioned on several fronts. On the DRAM side, we continue to be very pleased with the technology execution post the acquisition. Some really great results. I'll go into some detail on these for you on our 20-nanometer technology rollout, both in Hiroshima and in Inotera, both doing really well. In addition, we've now got what we call our 1X-nanometer technology moved from Boise over to be focused on in Hiroshima.
Our development model is working as planned and letting us focus more on the future. On the NAND side, good progress on our 16-nanometer TLC, which is our last planar node. I'll talk quite a bit about the decisions we've made and the benefits of the 16-nanometer TLC planar to vertical transition that we're going to be going through. We're very pleased with the progress on our 3D NAND. Right now, we've got big focus this year on, of course, our 3D NAND rollout that we've talked about on generally the same timeframe we've been referring to for the past year or more. Also now our focus shifting over on the development side to future generations and establishing a long-term roadmap on 3D. Good progress there.
On the package technology front, really a year of both continued maturing of the 3D packaging technology that is our HMC products that we're delivering to customers now. In addition, strong focus on the R&D package technology side on enabling the next generation, which is really coupled with our 20-nanometer DRAM products for even better performance in the future. I'll give you some additional detail. A lot of activity in the new memory front, as Mark referred to. I think we continue to be uniquely positioned with great partners and really looking at some diversified applications in the future. I'll talk about where we're going with our new memory technology. I wanted to start out today with a little bit of detail.
Mark referred to the complexity of DRAM. We often refer to complexity in moving to 20-nanometer and beyond. The impact that that has on the business overall. I wanted to try to quantify a little bit what that looks like. On this particular graph, what we've done is compared a node from seven or eight years ago, and at that time, it gave us 100% bit increase per wafer. It's actually a transition from 50-nanometer to 30-nanometer. Look at the process complexity of that relative to today's world, where we look at what it takes to go to 20-nanometer. In this case, I picked 30-nanometer to 20-nanometer because it gives you effectively the same kind of bit increase. In both these cases, you get 100% bit increase on a wafer.
If you look at the historical difference, many times we talk about increasing mask levels, and we quantify that. Certainly, you can see the difference between the node seven or eight years ago and today, where we had about 10% more masks, now looking at 35% between these two particular nodes. It's even more striking when you say, okay, we've added more masks, but really the process complexity in between these masks in order to enable the node and the process control that we need for that node to yield at high volume is really exploding also. More than 110% increase in steps between those mask levels.
These two things lead, of course, to the things we have talked about at a high level, which are, it takes more clean room space to get that 100% more bits today than it did seven or eight years ago, substantially more. Of course, that scales with number of process step scales with CapEx and how many you get out of a fab. Generally, message here on just trying to quantify what we mean when we talk about process complexity. Now I want to shift over to more on technology execution. As I mentioned, we're very pleased with the way 20 nanometers has gone. We had a big endeavor with the merger of Elpida and Micron, putting the technology teams together and putting together a roadmap to go execute on DRAM, put ourselves in a strong position in the future.
On this particular graph, what I'm showing is effectively the time from start in manufacturing to how the yield ramp looks through a period of time. It starts in the same place for all these nodes. As you can see, we set a pretty aggressive plan, which is the blue dotted line on 20 nanometer. We had good confidence in these teams coming together. We set a very aggressive plan relative to what we'd historically done on 30 nanometer and on 25 nanometer. To date, we've exceeded that plan considerably. When we say we're very pleased with what's going on on 20 nanometer in Hiroshima and in Inotera, this is the kind of thing that we look at that drives our enthusiasm for the execution of the team. Continuing to look now at, as I mentioned, we put 1X nanometer into Hiroshima.
We've got good focus on that, we're gaining confidence on additional nodes beyond 1X and 1Y. I think we're getting a lot more confidence in the ability to scale DRAM farther out into the future. On the 3D NAND side, a couple of points here. Certainly, we view our execution on 3D NAND really to be hitting right at the plan that we've had. Clearly, there's lots of discussion about how far you can extend planar NAND, and I'll talk a little bit about the performance trade-off you have when you scale planar NAND. I wanted to look a little bit about what we see as the change in bits per wafer and the cost structure over time between planar and 3D NAND.
A couple of things here. Hopefully out of this you'll see why we feel confident that our transition from 16-nanometer TLC to 3D NAND was the right time for us to shift all of our resources and put them firmly behind the 3D NAND direction. When we look at the bits per wafer, you can see where a 32-tier MLC part falls here relative to a scaled planar node. You can see there's some advantage, but it's not a huge difference in bits per wafer. We've talked before about the cost and the wafer trade-off, how many wafers you get out of a given fab for vertical relative to TLC. On here, you can see it's really a story of where this goes over time now.
It starts diverging significantly with significantly more cost or bits per wafer over time as you go to 32-tier TLC. Once you start going beyond that, you start getting good separation in bits per wafer. On the cost side, it takes a little bit longer for it to catch up because of the challenges we've talked about on the space and the capital involved. Ultimately, for a cost direction here, we view this as more of a historical cost per bit, ultimately scaling relative to what we had on planar over the past 10 years. Generally, the cost story on 3D NAND continues to build and continues to get better. I want to talk now on the next slide a little bit about the performance side. Without going into great technical detail here, fundamentally, you can think of the metric here, which is electrons per cell.
Fundamentally, you can think of that as really the core capability of the memory that the other capabilities are built around. When you talk about endurance and speed and these things, first off, you have to have a certain number of electrons per cell. The things to note on this are through time on planar NAND, the performance of the NAND has continued to degrade at a fundamental level. We've managed that through controllers and error management, but it continues to get more and more complex from a managing the performance point of view. The big disruption on 3D NAND, is it really steps back to something between a 50 and a 70 nanometer type technology for raw performance of the NAND. This is a big advantage for 3D NAND.
The other thing to note on this that turns into really performance for long term is as you scale the path by going to more and more tiers of vertical NAND, the raw performance of the cell capability isn't degrading any longer the way that the planar did. You have a different kind of battle here. We view this as a real long-term performance benefit for 3D NAND as well. Next, I'm going to shift over to future memory technology and kind of frame this up in the way we look at things. On the roadmap enablement piece, we look at our DRAM roadmap, our NAND roadmap. Then what else there is out there. On DRAM, as I mentioned, we're getting more confident in multiple nodes of scaling right now than we have been over the past couple of years.
We still view an opportunity for intersection at some point in the future with a type of memory that can extend the high-performance spectrum or end of the memory spectrum the DRAM holds into the future with new memory technology that fills that space. Part of our new memory focus is clearly around DRAM performance type enablement. On the other end of the spectrum, on the NAND side, we really don't view a technology need or one that we see out there that's capable right now of fundamentally being a NAND replacement. Part of that is because of our confidence level in 3D NAND, and we see it extending for a long time at a very good cost improvement basis, with really strong performance on just the pure storage side. We think that's covered.
What we see is opportunity in the middle for what we call storage class memory. Brian's going to talk some more about a lot of the applications this helps drive. Many of these are very customer specific, and we think this is a real strength of Micron. We have good partners, we have good customers that want to work with us on this to go enable a type of memory that has a performance and a cost position relative to NAND and DRAM kind of in the middle. Better performance than NAND, but better cost than DRAM. We think there's a lot of opportunity in there for things all the way from mobile to data centers, and I think that those kind of applications require close partnering with key customers, and we think we're in good position on those.
On the base technology to enable some of, especially the storage class, we're focused on things like spin torque and resistive RAM, and others that we're not talking about today that we think enable some of these storage class applications. Middle of last year, we had an announcement on a 16 gig RRAM, which is a kind of storage class memory. These are nice milestones on the way to actually having something that's commercially viable, and we think we're headed in the right direction on that. Last, I just want to finish up with a little picture of what we're calling our innovation roadmap. We look at this, and I think the notable thing is really this overlap of big technology execution programs in 2015, and Mark referred to that earlier. I'm sure it'll get brought up a lot today.
There's a whole bunch of critical technology milestones. I think we're really well-placed coming across these. On 20 nanometer DRAM, on 3D NAND, the rollout of at least the development effort on the next generation of Hybrid Memory Cube, then some things that we'll be talking about on new memory with probably a lot more detail, starting sometime second half of this year. We're feeling very bullish about our technology efforts and what we've accomplished over the last year, and we think things are looking pretty bright. That's where I'm going to finish.
Hi, Scott. Srini from Summit Research.
Yeah.
Question is, what node would be the 3D NAND at? Will it be 40 nanometers or 50 nanometers or 30 nanometers?
Sure. Our technology node on 3D NAND is a little bit complicated to talk about. For example, some of our competitors talk about a technology node in terms of really the thickness of a film that's in their stack, which ultimately is a feature of the memory, but it really isn't very relevant. It doesn't change the bit density, whether you're at that 30 nanometer or if you're at 20 or if you're at 40, the bit density is still the same. Ultimately, what we look at is not technology node. Well, we have litho levels that are immersion on our 3D NAND that are probably similar to what some of our competitors do. We look more at optimizing really bits per area. We will be very competitive on our bits per area with anything that's in the market today.
30 nanometers of film thickness, which I don't want to talk about today of what our film thickness is going to be, but I will say it doesn't mean as much as it's being played out to me. That one's tough for me to answer, too. It does if you are running Yeah, it does. Our technology will have the similar number of electrons on it to what other competitor technologies do. Okay. We'll be able to answer that exact question for you a lot better here over the next couple of months when we're a little more public about what our technology is.
Scott, here.
Sorry, I can't see. Yeah.
I can see this. You had the slide about technology roadmap, especially around DRAM, 20 nanometer all the way down to 1Z. Earlier on, you also had a slide talking about increased number of steps, especially with non-litho and how the clean room space changes. Can we take some of that information and try to overlay it as you go from 20 to 1X and 1Y? I'm just trying to better understand how complexity changes and how much of a clean room you have to give up.
Well, the trend on DRAM for increasing complexity is, yes, it absolutely continues, a lot of that is driven, I think, probably as you know, by the number of multiple patterning levels you have to put into the process flow, which does continue to increase as you go to 1X and 1Y. I think the trend is similar for the next node relative to what I showed you earlier. It's another pretty substantial step to get another 100%. To get that complexity down, if you go to something like EUV, certainly we'd be looking at something like that over time to help with that, it doesn't completely solve it either. Part of what Mark alluded to earlier is we just have to be very cognizant of the trade-offs we're making.
We look at a new technology node and we're very aware of how much complexity we're adding versus how much benefit we're getting. Of course, we do the node only when we get that benefit, but it's more complicated than it used to be just to do a scale and have it make sense. I think the cost curve, you mean the cost reduction per node? Yeah. Is absolutely a challenge. You wouldn't do a node if you didn't get a cost improvement that was substantial, but it's more challenging to get the benefit of a given node. Without a graph, it's hard to talk to exactly what it is, but absolutely, it's challenging to look at the complexity we're bringing in and then make sure that you've aligned everything so that you actually get a cost reduction with the next node. Performance is a different thing sometimes.
On the cost side, it is more challenging now than it was 10 years ago to know that we should do another node. I'm sorry.
Scott, can you repeat that question?
It seems like the process recipe for 1X and 1Y may not been finalized. This is why you don't have the slide to be able to better assess the increased complexity. We're still working on finalizing.
Well, that's historically always been true. What we do is project off of You never know what the exact process recipe is three years from now on a given node. That's absolutely true. We have pretty good projection for what the raw process time changes are on given process steps that we know are going to be in the flow, and we know how much it's going to increase. We actually have pretty good projections for how much complexity is going to come in on a 1Y or a 1Z node right now.
You're not going to share that.
I'm not sharing them yet, yeah.
Harlan Sur with JPMorgan. Scott, thanks for the presentation. I got two questions. Number one is, when you talk about 3D introduction this year, it's 32 layer. Are you guys planning to roll out two bits per cell or three bits per cell? Where's the focus going to be this year?
Sure. Our focus is absolutely on three bits per cell, but we will be rolling out some products are better suited for MLC, depending on the exact performance and density they need. We'll have MLC products, but absolutely, we are very focused on TLC coming out really early.
Great. Obviously, you guys talk a lot, and your peers talk about the complexities associated with 3D NAND. We hear about yield issues. At the end of the day, it boils down to manufacturability, and it boils down to, can you continue to drive costs for bit scaling? The question that I have for you, because it sounds like your 3D NAND process is already finalized and you guys are working on manufacturability. Help us understand cost per bit 3D NAND TLC relative to your 16 nanometer planar TLC products. Are you guys at cost per bit parity yet or not?
We are going to be at cost per bit parity later this year, most likely. We're still in an early phase, as Mark and others have mentioned, and I think we had on a slide. The real ramp in Singapore is going to start this summer, and we'll be ramping it through the year. Always when you're first starting up, it's actually a little bit hard to get parity until you have enough wafers running in the fab to have a real cost basis for something. Our 32-tier TLC absolutely will be at a better cost structure than our 16 nanometer TLC. This graph kind of looks at where, at the right yield levels, the difference between what our 16 nanometer TLC would be and what our 32 is.
Thank you.
Hi. Thanks for the question. This is the same chart. If you look, seems like 32-tier TLC is not giving you much cost advantage with your 16 nanometer TLC. That means once you do that node transitions, that means you don't get first-year much cost decline as you transition from planar 16 nanometer to 32-tier TLC?
I'm not sure I caught the whole question there.
If you draw the line, 32-tier TLC seems like very similar cost with your 16 nanometer TLC.
Yes
right?
It's a little bit lower, but yes. That was a comment I made earlier, is really you get some cost advantage on the 32-tier relative to 16, but it's really a strategic direction that you know you're headed down a path where the cost advantage continues to get better and better over time.
That means you will just do some 32-tier and then move to 48-tier fast so that you get the cost benefit?
We will move relatively quickly to a next node, although there is a cost advantage for 32-tier. The second point is there's an absolute performance advantage. This is both a combination of getting the cost structure on the right path, as well as enabling really some key products that Darren and Brian and others will be talking about through the day with the performance of 3D NAND. There's two benefits.
Hey, Scott. You showed a chart showing that 20 nanometer yields were progressing faster than prior generation nodes. That's a little counterintuitive given all the discussion on increased complexity. How should we think of maximum yields at 20 nanometers compared to previous technologies, just given the complexity? Just help us understand why 20 nanometers can be better than prior nodes across a number of these metrics.
Sure. Well, I can tell you from a Micron point of view, I think we do see mature yields that are very similar or I want to say better, but at first case, I think it's at least similar to what we've seen historically. As Mark talked about investments we're making through the Elpida acquisition and moving forward, we've invested a significant amount of resource that neither of the two companies by themselves had in terms of process capability and ultimately yield capability focus, just with the bandwidth we're applying to this right now. Absolutely, if we'd had that same bandwidth through history, then 20 nanometer for us would be challenged relative to some of those other nodes. Really, we kind of went through a reset in how we do DRAM development and the kind of focus we're putting on it over the last couple of years.
We're pretty bullish on where 20 goes relative to anything we've done before.
Hey, Scott. Vijay from StoneAge. Just a question on 3D NAND. When you look at 3D NAND, it's pretty high fixed cost, right? Your output on the fab goes down to 50%, around there. You have some increase in bit growth. When you talk about cost parity to 16 nanometer, what kind of wafer output are you talking at to get to that cost parity with 3D versus 16 nanometer?
One of the points on this one is the bits per wafer on this 3D NAND node are actually substantially higher than 16. You just don't get that all passing straight down to cost. You're going to have to be in sub 3,000 or 4,000 wafers a week when you start seeing what your actual costs on 3D are, and then the cost comparison's more valid. When you're running small numbers of wafers, the cost comparison is just not very valid.
I see.
As soon as we have a stable line ramping up through the second part of this year, then it won't be too long until we hit good cost.
Can you talk about the capital spending ramifications of this 3D transition? I think you're spending less than $1 billion on NAND this year. How much does that buy you? Then what are the economics of sort of adding 3D capacity versus converting planar capacity to 3D?
I'll probably let Mark cover a few of those at the end of the day. I know he's going to have some comments relative to that capacity overall. We are spending a lot of time looking at, as we've talked about, adding some white space. We have some white space in Singapore now. Then we'll do a careful view of when it makes sense to convert planar over to 3D NAND. I think I'll punt that one over to Mark a little bit to let him talk about that one. Yeah.
Hi, Karl Ackerman from Cowen and Company. Relative to 3D NAND, I think Mark was saying that 3D does not necessarily add to bit growth absent wafer capacity adds, I think everybody's clearly going to add 3D wafers. I guess the question is, how much do you think 2D will ultimately be converted to 3D?
I can't speak for our competitors' strategies. I can say from our point of view, we are going to convert some of our planar NAND over to 3D if we look at a site like Singapore. At the same time, we have other important businesses that depend on high-performance planar NAND nodes that we're going to support for a significant period of time into the future. Jeff will talk about some of the embedded opportunities, different things there that are good business for us to continue to drive planar wafers. We're not going to convert all of our planar wafers to 3D NAND, but we're absolutely going to convert some of them.
Question on EUV. Can you talk about where it is on your roadmap today, and how many layers you would expect to use initially, and how you think about cost of ownership there in terms of wider adoption?
Sure. Right now, there's been pretty good progress on EUV over the past year. Again, it's still not quite at the position where it makes sense for Micron. If you look at fundamentally where it comes in, most likely you'd bring it only in for a couple of layers to start with, but ultimately it could replace 10 or 15 of the levels if it was at a much higher throughput than it is today and reliability point. One of the things that when we look at EUV that's sometimes lost is by the time it comes in at that throughput point, it's going to be multiple pattern too. EUV is not coming in and we're just going to do a single print in very many applications once it comes in.
We're still going to be doing multiple patterning, maybe just two multiple pitch doubling instead of pitch quad. It's not coming in as an absolute cost saver. It's complicated to bring in. It's also complicated when you've already got fabs set up to run double patterning and quadruple patterning with equipment that's been in there for a period of time. EUV's got a pretty high bar on something like DRAM or NAND, where the fab's already built.
One more question, Scott.
Okay.
Great. Thanks for taking my question, Scott. Doug Freedman, RBC. When I look at your charts, the one that you're showing up there now, it shows a planar roadmap with 16 going down to a 1Z and then a 1Z prime. If I compare that to your innovation roadmap, Micron, you're showing your roadmap stopping at 16 planar.
Yes.
Am I reading that correctly? Is there no more scaling for you?
You are reading it correctly, Doug. The point on this is our projections, which are the line. I'm not saying what we're doing on this node on this particular graph. I'm saying what our technology cost and bits, if we do any of these nodes, would be. It's a projection graph. For Micron specifically, we're stopping at 16 nanometer TLC, which is before the last two dots on there, which are what we could conceive of somebody else deciding they might want to do.
Okay, great. You believe that the industry could have another planar step beyond the 16 going
Yeah
to 15 and beyond?
Yeah, absolutely.
Okay.
The industry can absolutely continue to scale planar NAND. If you look at the reason why we've chosen not to is more about this one, and it's about performance. If you're fundamentally looking at a technology that has hundreds of electrons to deal with and try to make a TLC part of high quality in the future versus one that has six per state, and you're trying to keep track of somewhere between six and 10 electrons to make a planar NAND work, you can make it work, but you have to deal with different kinds of problems, and the quality levels is a challenge. Some applications, that may be fine for. Okay? All right. I think I'm done, right, Ivan? I don't see you. Okay, thank you.
Okay. How's that? Good morning. I thought I'd start off with a very quick refresher on what CNBU is. We have responsibility for selling the Micron portfolio into PC, that includes graphics, into servers, both in enterprise and the cloud side of that, as well as into networking, all with the exception of storage. SSDs and NAND that goes into storage, and you'll be hearing about that a little bit later on from Darren Thomas. That's managed by our Storage Business Unit. In talking about the opportunities that we're looking at, there really are three basic messages that I'd ask you to take away.
The first is around the sources of opportunity for growth in our business going forward, and the fact that there is an ongoing and continued shift away from PCs, which has been the driver both for the company and for the Compute Business Unit for quite a while, towards the incredible growth driven in the enterprise and the data center. The second is the ongoing continued growth in the richness of mix of the products that we are selling. With the exception of networking, which has been a very complex product mix for some time, if you go back a few years, these segments could essentially all be serviced with a single product, DDR3. If you look at the transitions underway in client, in the thin and light space, it's moving towards LP. In graphics, it's moving towards a very high bandwidth specialized graphics technology.
Servers were transitioning over to DDR4. I'll talk about some of the unique technologies that we're developing like RL and HMC. Again, it's a much richer mix of products which we think makes for a little bit stickier business, but it also is one of the drivers for the operational flexibility that Mark talked about at the beginning, and we just got to get better at that to be servicing that more complex mix. The third thing I'll talk about, though, is really building on what Scott has just discussed, which is the increasing opportunities to leverage our technology investments. Of course, part of that is the traditional technology evolution, building out our transition, completing our transition to 25 nanometer this year, enabling and beginning the ramp to 20 nanometer.
Just as exciting are leveraging some of the packaging technologies for unique value-added solutions that we think can provide a better return for the capacity and for the investments that we are making. Let's quantify a little bit of what I was talking about with regard to the shift in the drivers of growth. I think it's been fairly well understood, the relative growth between PC and mobile, the fact the market opportunity is larger in mobile this year than it is in PC, and that gap is going to continue to grow forever. Mike Rayfield will talk to you after the break about some of the incredible opportunities that presents for us on the mobile side. Perhaps a little bit less well understood is within that compute market, how the growth is getting driven in the enterprise and the cloud.
I'll talk in a bit more detail about what some of those specific drivers are, but it's driving in the 40s in bit growth rate on the enterprise side, in the 50s on the cloud side. These are compound growth out through 2018, to a point where in 2017, the combination of enterprise and cloud will pass PC, and by the end of the decade, each of those segments individually will be larger than our opportunity in the PC space. Networking, obviously, it's a much more modest size, but it's much more diversified. It also demands longevity, makes it much stickier and a good value opportunity for us. The growth of video traffic generally drives a very healthy growth rate there.
While PC is modest in growth, high single digits, it is, we've got to keep in mind, it's larger than the rest of our bit opportunities combined this year and will still be the largest segment for us this year and next, obviously an important one to keep an eye on. Then graphics is a much more modest growth, although the shift towards a proprietary interface, we think, provides a very good value opportunity. Now let me shift and talk in a little bit more detail about each of these markets. I'm actually going to go from the lowest bit growth opportunities to the highest, starting with graphics. I mentioned that it is a fairly modest bit growth opportunity.
While there is demand coming from 4K and ultra-high density, as well as the need for ever more realistic gaming opportunities, there's also a lot of pressure coming from the ongoing switch towards mobile platforms for people to play games on. That's what keeps the bit growth modest. Why we're interested in it is that this drive towards higher resolution, more realistic gaming environments, is driving the need for very high bandwidth. We're developing and shipping a technology called GDDR5, that operates at 8 gigabits a second, exactly the fastest interface of any traditional DRAM component on the marketplace today. At those kind of speeds, it requires a very intimate cooperation with the chipset partner that is developing the chipsets, as well as with the customer that's developing the boards, and that makes it quite sticky.
Therefore, we think it's a good margin opportunity for us. From a raw technology perspective, it also happens to be fairly friendly to a new technology. It's relatively easier to ramp than in a market like server. It'll actually be the second place that we ramp 20 nanometer later this year after PC. We see, through continued focus on our technology and continued very intimate cooperation with our chipset partners, that we can continue to drive this bandwidth even higher. We see a chance to double it again to about 15 gigabits per second going forward, which will certainly have opportunity in graphics and may have opportunities in other market segments as well. OK. I did talk a bit about how the growth opportunity is really moving away from PC.
Again, keep in mind it is our largest bit opportunity within the compute BU this year and next. So it's obviously very important. There is a degree of product portfolio diversification ongoing here as well. Certainly, over time, probably start next year, we'll see mainstream desktops kick over to DDR4. We're also seeing, driven by battery life and power, a shift towards LP technology, initially LP3, moving to LP4 in the thin and light segment of notebooks, again, contributing to that richer product mix that we need to be managing. Also, in those thin and lights, typically that LP component is soldered down on the board, and again, it makes for a tighter, a little stickier relationship in terms of how we supply. From a technology perspective, this is the first place that we will be enabling and ramping 20 nanometer.
As Scott said, we're very pleased with where that's going, and it is a PC product, 4 gigabit DDR3 that's on the lead there. So we're going to start production of that next quarter, and you'll see in the latter half of the year, this starts to become a significant contributor to our business. Networking, second-largest growth opportunity for us, and the growth opportunity here is being driven, in a word, by video. Whether that is the user-generated video like YouTube, whether that is the increasing growth of over-the-air delivery of traditional content from Hulu, from Netflix, even traditional networks like ESPN from the likes of Dish. That is driving a very significant growth opportunity in the infrastructure as well as in LTE wireless deployment to deliver all of that bandwidth to the users.
From an innovation point of view, this is an area where we've been doing segment-specific innovation for some time. The segment is very focused on low-latency products. So we've had a family of reduced latency DRAM products, the latest being a so-called RL3 product that is being very well received, a very good high-value product because of its low latency. This is also one of the areas where we are leveraging not just the raw technology evolution that Scott talked about, but some of the packaging and 3D technology as well.
As some of you may be aware, we have been for some time developing a technology called Hybrid Memory Cube, which takes basic DRAM technology, interconnects that technology with something called through-silicon vias, which allows power and performance across chips similar to what you traditionally could do on-chip, and then connects a stack of DRAM to a control layer, which by managing a very high-bandwidth interconnect to those DRAM through the TSVs, you can manage error correction, repair at a very high-speed serial interface. It allows for higher density, lower power, and lower latency at an overall solution level than you can achieve with standalone components. One of the areas that is very receptive to this high bandwidth, this low latency, is in the networking space. As we begin to ship qualified product later this year, will start to become a contributor later this year and next year.
Again, in a very high-value segment because of the value that the networking space places on this. I think the last comment I'll make here before I move on is this is probably the one area, more than others, where longevity is a value that is appreciated by the customers. We take advantage of some of the work that Jeff Bader does in his embedded group, focused on automotive and industrial markets, with things like a Product Longevity Program, because these customers will be selling into operators that really don't want to have to do rapid forklift upgrades. We can get a good premium as well from our ability and willingness to be providing technology on an extended basis. I mentioned, of course, the shift in the growth drivers going over to servers, going into enterprise and cloud.
It's really coming from a number of factors, but one of the most significant is a growth of what we refer to as high-velocity applications. We think about that both in terms of so-called structured database applications. Think of things like SAP's HANA in-memory database or Oracle's TimesTen, where the performance demands on those traditional databases are getting to a point where they can't afford to be paging off of disks or SSDs. They need to be operating out of memory. That is driving demand for basically as much memory as you can possibly pack into a single server. Similarly, in the so-called unstructured world, think of problems like graph analytics, which might be a social network trying to map the connections between their users to generate the best advertising.
It might be an insurance company that has to come up with a real-time insurance quote, so again, using graph analytics to come up with a risk profile so that they can do that effectively, again, driving tremendous growth for density of DRAM in a box. If you look at the server business across both of those segments, the unit volume of servers is actually not growing that much. We anticipate it's in the single-digit range. But the content per box is growing in the 30s, perhaps even pushing low 40s. That's what's driving that growth in enterprise and cloud up to the mid-40s and in enterprise, and up to the mid-50s on the cloud side of things.
How we're responding to that, certainly at a very, perhaps most basic level, certainly managing the transition to DDR4, which is going to occur initially in the enterprise space. The cloud players generally migrate a little bit slower. That does provide power benefits to the user. From our perspective, that transition actually takes a little bit of bit capacity out of the market because we do lose about 10%, maybe a little bit more, bits as we go from DDR3 to DDR4. Of course, from a technology perspective, we're leveraging the advances there. We're completing our transition to 25 nanometer, and we're in the process of enabling 20 nanometer in this space, something that'll be happening from a ramp perspective towards the end of the year.
I think a little bit more exciting in terms of some of the specific opportunities for higher density, I talked about this interconnect technology called Through-Silicon Via, which allows very high-performance connections across die. That is one of the ways that we will satisfy this thirst for ever-increasing density per server box while still maintaining the increase in performance. You're seeing server platforms now moving to 2.4 gigahertz and beyond. To get that performance level with a stacked solution demands TSV. The traditional stacking technologies just won't get you there. We're looking at how we move from a mainstream 16-gigabyte server DIMM today, and with TSV technology, moving up to 64 and even 128 gigabyte per DIMM, again, satisfying that nearly insatiable demand for density in server boxes.
The other one, again, similar to the networking space, in the compute world, the piece of this marketplace that has that need for extremely high bandwidth, extremely low latency, managed in a power-efficient way, is the high-performance computing space. So this is the other segment where we're seeing early uptake of Hybrid Memory Cube. Again, we'll be shipping qualified components later in this fiscal year, and beginning that ramp, and we'll see that again towards the end of this fiscal year and into fiscal 2016, become a significant contributor of high-value revenue for us in this space. Of course, going forward, as Scott mentioned, we are very much focused on the next generation of Hybrid Memory Cube, which doubles the density, doubles the bandwidth, and again, in both the high-performance computing and networking space, we think will allow us to continue to grow very well.
If I could kind of summarize where we are focused and taking advantage of these opportunities, certainly execution, just basic blocking and tackling is very important. That is working closely with Scott's team and making sure from an execution perspective, again, finishing up the 25-nanometer conversion, certainly in the server space, enabling and beginning the ramp of 20 nanometer, and staying very focused on the operational flexibility that Mark talked about at the opening, because this complexity of mix demands that we have very good agility to shuttle capacity back and forth, obviously between segments in CNBU, but also with the other groups that use quite a bit of DRAM, mobile, and embedded as well. A critical focus on that, of course, while continuing to drive very high levels of quality and service.
Thinking about a way to frame many of the value-added solutions that we are working on, when you look out broadly at a lot of the compute problems that we are trying to solve going forward, the energy and performance associated with how data is moved is becoming as important, if not more important, as a factor as how that data is being processed. If you think about a number of the evolutions in memory interface that we're talking about, we're focused on reducing latency with products like RL3. We're focused on higher bandwidth with products like GDDR5. We're focused on lower power with our LP products and the move to DDR4. We're focused on all of those with products like HMC. Again, all with the eye towards managing that movement of data.
In fact, if you think about the ultra-high-density drivers in the server space, one of the drivers for things like SAP HANA and Oracle TimesTen is to manage, that is to say, to reduce, the data movement, in this case, not between the processor and memory, but between memory and the storage device, be that SSD or rotating, and dramatically reduce that churn, which drives growth. I think going forward, beyond just focusing on how we manage those optimized interfaces between the processor and memory, between the memory and storage, we also see and begin to see emerging opportunities to put very basic elements of processing into the memory itself. Of course, as you can do that, you can further reduce the need for that movement of data. One of those opportunities, of course, is the Automata Processor, which we've announced.
Brian Shirley is going to be talking about that as he takes a little bit more forward-looking view about what we're doing across all of the business units. With that, again, repeat Mark's thanks for joining us here, and would like to turn things over for Q&A.
Yeah. Over here. Thanks.
Yeah.
Daniel O'Meara from Lam Reasearch. In terms of the 3D NAND and 20-nanometer DRAM and 3D NAND, can you highlight a bit how that fits into the product portfolio here going forward? Does that change anything from a competitive perspective in what you can offer these customers now?
Well, with regard to 3D NAND, that really is aimed at our storage portfolio. I'm going to defer that question to Darren Thomas as he gets up and talks about SBU. From a 20 nanometer perspective, over time, that becomes relevant to all of the segments that we've talked about, just in terms of the nature of the technology and how it evolves, as well as the process of getting it designed in. That will hit first in PCs, be followed very closely thereafter by our graphics products, then, a little bit further out as we drive to the necessary quality and reliability levels, that'll hit server. Probably the slowest to move tends to be the networking players.
They want you to keep building the same stuff for upwards of five years, the flip side of that is they tend to be a little bit slower to adopt the new technologies.
That's also, I just point out, Tom, obviously mobile as well.
Oh, of course. Yeah.
I don't know if this question was.
Absolutely. Yeah. Thank you.
Tom Eby, in Mark's opening remarks, he mentioned how the company needs to transform to where you are working more closely with your partner to help them differentiate.
Yet, that wasn't a real part of your presentation, and I don't know that it's really been a part of the company's history, because customer just wanted a lower price historically. How do you make that transition? This seems like it's very difficult and yet could be very meaningful.
I think that there's one fairly basic level, which is, as you move from JEDEC-driven, more standard products towards products that may, at some level, be JEDEC, but are more differentiated. There's a much tighter cooperation with the ecosystem, with chipset partners and others. A couple examples, I talked about the very high-performance graphics products. That ends up being a very tight relationship between the graphics processing companies to make sure that interface is tuned and working well, even to exactly how the customer lays out the board, because how you lay out the traces there is important. That's one area. I think HMC is another good example where, again, both with the networking community and with the high-performance computing community, we have a very tight discussion with them well in advance.
Right now we're talking a lot about the Gen 3 HMC to understand exactly what their workloads are, what their compute problems are, and how we can be tuning our solution, particularly from the controller and the firmware level, to make that work. I think, in terms of how we help drive that transition, there's a whole bunch of things that we're doing, but certainly part of it is the team and how we augment that. Mark had listed some of the people that are new to the company, and I'll just touch on a couple of them that are within CNBU. At a very high level, I think we have long had many people that do a superb job that think about the system from the perspective of memory.
What we're trying to do increasingly now is change that and say, "Hey, let me think about memory from the perspective of the system, and what can we be doing in driving the memory technology and the interfaces so that it could be a more valuable and helpful part of that system?" The guy that's driving Automata and some of the advanced work that we're doing, again, to try and take advantage of reducing this data movement problem, a guy named Stephen Pawlowski. He joined us about seven months ago. He had a long career at Intel in various roles driving CPU and platform architecture. That's somebody I think that brings a different perspective. More recently, we hired a gentleman, Bob Quinn, who had actually founded a company called 3Leaf, which is a very innovative processing and node interconnect company that got sold to Huawei.
In addition to understanding the compute and networking space, actually spent three years working for Huawei, a very useful perspective on the China business. Actually before he joined us, was chief technologist at LSI, looking at storage problems. In this infrastructure space, it really is a convergence of compute, networking, and storage. While we're very clear about how we manage that from a BU perspective, bridging those connections across Darren Thomas' and my groups is important. Just a couple examples of some of the people talent that we're bringing in to help continue to drive that change.
Tom, hi. It's John Pitzer with Credit Suisse. I was hoping you'd talk a little bit more about the high-velocity applications that you talked about driving densities within enterprise and cloud.
How should we think about the densities of DRAM today? Where will it go? What % of the market is being driven by high-velocity applications? There's a debate in the investment community as to where compute stops and storage begins. How do we think about the DRAM in that system versus a NAND in that system as you start talking about things about big data, in-memory database, and things like that?
In terms of the opportunity, we'll go back. This is from a DRAM bits perspective, looking at where the opportunity is. When you get out to 2018, again, you're almost twice PC in that enterprise space. If you were to look at this for a DRAM in total, this would be, again, in 2018, roughly 30% of the opportunity being in this enterprise and cloud space. The largest piece, actually, I think mobile, if I recall, is about 40%. Those become the two big drivers. From a growth perspective, it kind of steps into second place as we go forward in terms of driving overall bits in the industry behind mobile. Again, I think it really is a drive for as much density as we can put into a single box, onto a single DIMM, at some reasonable premium, right?
There actually are fairly healthy double-digit premiums per gigabit today when you move above the mainstream. Today, that mainstream density is 16 gigabit. You move to 32, you move to 64, you're getting a nice uptick. For us, it's about how can we drive that density without taking undue cost penalties. We're currently thinking we'll do four high TSV stacks on eight-gigabit components, which will enable 128-gigabyte DIMMs. We'll keep looking at, hey, is it economical to be driving higher stacks, the DDR4E, the next generation of technology. It looks like it's going to get standardized at eight, 12, and 16-gigabit densities. As Scott keeps doing his job on technology, we'll keep driving that. In terms of how to think about DRAM versus storage, I think that's a fairly well-understood division today.
Obviously, there's a steady and very rapidly growing shift from rotating to SSDs, and the place that's happening first and foremost are in these high-velocity applications, where they value that lower latency access. My personal sense, and again, I'll defer to Brian to talk about this in a bit more detail, is that some of the exciting evolutions there are the potential to further disintermediate that memory hierarchy, where way back when, it was just a processor, DRAM, and disk. Now we've inserted a lot of layers of cache and SSDs in between, and some of those emerging technologies are a further disintermediation between DRAM and what is NAND today. That's a good opportunity for us. In terms of more specifics about that, I'll let Brian talk about that later this morning.
Tom?
Okay.
Thanks. Excuse me. Steven Fox, Cross Research. Can you just talk a little bit more specifically about DDR4 and where you see it ramping as you maybe exit the year? What % of your shipments it could be, what's the most obvious applications to have success with?
Sure.
Also compare it to where it's going to exit, maybe in 2016, as a % of your business.
Yeah. Okay. In terms of where it deploys, the first place it deploys is in the enterprise side of servers. That transition is underway today, and that's going to be what drives the bulk of the transition this year. I would expect that as we get out towards the end of the year, we won't have gotten to half of our enterprise, to half of our server bits being there, but we'll be approaching that. The second place that that transition occurs will be in the cloud. Those folks have a little bit more sensitivity on the cost per bit delta there. It's going to come, it's just going to come there a little bit later. As we get out into 2016, you'll start to see the client platforms transition.
As we get through 2016, there will start to be a fairly significant shift over in client. Certainly, by the end of 2016, on the enterprise and cloud side, we will be very heavily converted over.
Okay.
Okay. Very good. Well, I think that is what we have time for, and I am going to turn things over to Jeff Bader, who will talk to you about our embedded business.
Tom.
Thanks. We will follow a similar format that Tom did in terms of giving you an idea of what the embedded business is about, where we are focused, and where we see the same growth drivers. Embedded for Micron means automotive, industrial, consumer, and what we call connected home. It is those application segments that we are focused on within the embedded segment. If you think about it, Tom mentioned a little bit some of the need for longevity, and they are sort of ordered in that order as well. You think about automotive design applications are driving, where they have very long longevity requirements. Industrial sort of falls somewhere in between. The consumer connected home part of our portfolio ends up tending to be closer to leading-edge technology and leading-edge transitions at the same time.
When we look at what is going on for us in the space, those three core markets, it is four segments, but if you think about it as automotive, industrial, and the consumer bunch, the automotive market is just, of course, growing very rapidly today, really driven by the amount of semiconductor and the amount of technology that is going into the automotive applications, specifically the infotainment application, and a growing portion of that going into what we call advanced driver assistance systems, or you will see it later as ADAS on the charts. That is really the story in automotive. The other story is the adoption of leading-edge technology there. Increasingly, in order to deliver the experience and deliver the safety feature or the security feature they are trying to deliver, it is requiring an adoption of very much leading-edge technology, which is a very new thing in automotive.
We'll talk about the things we're doing to help our automotive customers adopt that new technology. In the industrial space, it's really, in essence, the Internet of Things. It's a big buzzword, but what do we mean by that? Essentially, you have tremendous growth driven by the addition of connectivity and intelligence into what were traditionally unconnected and dumb devices, for lack of a better term. That intelligence and connectivity brings with it a compute and memory problem that we can solve from the Micron side. Essentially, it's distributing that data throughout the network. It's part of what drives this tremendous cloud and server growth that Tom just talked about. It's all of this data coming from these now distributed and connected devices in automotive and industrial. That connectivity, of course, continues over into the smart home and the applications in the consumer space.
In consumer and connected home, it's really the 4K transition, and in essence, the smart TV transition, often coupled with this 4K transition. We see that the price points on 4K are starting to reach a place where we believe that that transition is going to happen here. It's not only the TV part of it, but it's all of the content streams, essentially, that feed that TV. Think about your set-top box, what we call the over-the-top set-top box, or the Roku box, Fire TV kind of boxes. Delivering that IP connectivity and streaming into the home, delivering through your set-top box, all of those are now trying to go get ready and deliver a 4K experience as well to be paired with those. Both of those are driving significant bandwidth and density growth. Take a look at that growth a little bit.
Again, I spoke about automotive. You can see basically a 39% growth. We split out the NVM compound annual growth and the DRAM because we ship a fair amount of both in all of these applications. The growth in the data part, in the NVM, is really driven by things like that infotainment platform. Growth in map content, in navigation content, emerging sort of black box applications, those are all driving a significant growth in the NAND footprint that's going into these cars today. On the DRAM side, it's similarly driven by the features that are trying to be offered within that infotainment cluster. The other major driver of that is, as I mentioned, this ADAS systems or advanced driver assistance systems.
Think of that as your lane departure warning, your forward collision warning, your rear camera sensoring, the sort of what they call sensor fusion, so the integration of all of that data in real time to make a decision on autonomous or semi-autonomous driving. In industrial, I mentioned it's the IoT build-out, We think about this in, again, simple sort of unconnected devices now integrating in a full Linux stack, integrating in IP connectivity, integrating in, therefore, the NAND and the DRAM footprint required to go drive that and support that. In the consumer side, it's 4K, The network around that 4K is the big driver that we see here, again, driving a significant growth in DRAM per set in the TV case, going from a one or a two gig up to a four to six to eight gigabyte solution.
That's driving a tremendous growth there, as you can see from the CAGRs on the DRAM side. Now I'll dive into each one of these segments a little more, and what do we see as the requirements in this space, and what are we doing to leverage that and then to deliver a solution there. In automotive, historically, it's about automotive quality, first and foremost. These are the mission-critical applications in your automobile. You don't want a quality issue. You don't want a reliability problem. You have a very long life cycle. Tom mentioned before something we introduced about two, three years ago a program called the Product Longevity Program, which was essentially a select set of products that we made available for a 10-year form, fit, and function support.
That's made it very simple for our customers to adopt technology into these long life cycle applications. The increasing change in automotive is the adoption of new technology, adoption of leading-edge technology. We've been working very closely with Scott's team at the very upfront technology definition point to build in better support for the ultimate application in automotive. Just to put it in perspective, DDR2 adoption in automotive was basically about four years after adoption in PC. The eMMC, which is our mainstream managed NAND product that goes into automobiles, was about two to three years after adoption in cell phones and handsets. We're looking at an LP4 adoption today in automotive. We're going to begin sampling second half of this year, and we're going to be ramping in production a year and a half-ish after it ramps in cell phones.
A much quicker adoption of leading-edge technologies in order to solve essentially the compute problem that they're trying to solve in the car today. We're very much a leader today in automotive. We're about twice the size of our nearest competitor. Part of that is the long investment we put into this, building the custom solutions, building unique products, and supporting that customer base. We have a very strong portfolio across the space. The other major thing we've been doing for the last several years, there was a question at the end of Tom's thing about how are we working with customers to enable their innovation. We've invested strongly in a series of system engineering labs and customer validation labs that we've deployed now in multiple regions around the world.
The idea behind that is to engage directly with the car OEMs, to engage directly with our customers, to engage directly with the tier 1 suppliers, and the ecosystem and chipset suppliers. We're working with them in these labs to do usage model exploration, new technology exploration, essentially architecture development, to help them prepare for the adoption of this new technology. That's been a huge improvement to our relationship with car OEMs, and it's been a huge enabler of them being able to adopt this technology as fast as they're adopting today. In industrial, again, this is also a market that requires longevity, but this market in particular is a highly fragmented market.
When we go after this market, we leverage quite a bit tools and techniques to try to extend the reach to hit into more parts in this market rather than try to get to each customer alone. Again, in this market, you can see some of the applications down there, just a tremendous breadth of application space. It's sort of serving that breadth of application space that requires interesting and innovative ways to go reach the customer base. I'll talk about those on the right-hand side here. The other major trend, again, I mentioned connectivity and intelligence getting integrated into all of these different applications. One of the ways that we're engaging and sort of starting into that space is with the integration of wireless connectivity. Many of these applications are integrating essentially cellular connection.
We are a leader in that portion of the market. You can see it's projected to grow from around $350 million to $2 billion over the next several years, really driven by the 4G LTE build-out in these applications. Most of that market today has been the 2G, 3G networks, and they're converting to 4G and LTE, which is driving a much richer solution from our side. That's moving from a mid-density, low-density, NOR-based MCP solution to a two and four gigabit NAND LPDDR solution. It's a great opportunity for us to leverage the position we have today and continue to grow that. I've mentioned this fragmented market. It's really a key challenge, if you will, for reaching this market. Our focus there is, again, to work with a broad set of customers and a broad set of partners from an enablement perspective.
We're invested very heavily in the last two years on our distribution network, industrial distributors that we use, really figuring out how and where we partner with them to have them be the feet on the street and drive some of this demand creation, drive some of this business opportunity for us that we can then partner and help support. The other thing that we're working very clearly is with our chipset partners, and with the reference design partners. There's an example here on this picture is actually the Raspberry Pi Generation 2, which you may have heard they announced about two weeks ago, three weeks ago. It's a way for us to work with a customer like that who buys Micron, and then that design goes off to a number of embedded applications.
We're working with other sort of single board computer guys and reference design houses that we can get a design effort that we do multiplied out into this industrial segment. It's our key sort of strategy to go reach these kinds of markets without putting a tremendous amount of resources internally. Consumer and connected home, I put them on the same picture here. This is a more traditional technology, lithography generation, get to the latest technology type of market than either automobile or industrial is for us. It's about time to market. It's about early adoption of new technologies. The two big threads that we see here, I mentioned one of them before, is the 4K build-out, and it's sort of a 4K and all of the other streams that need to be 4K that are touching that.
We call it smart home in this space, but in essence, I think of it as the consumer version of Internet of Things, in the sense that various different places in your consumer, whether it's your smartphone, your wearables, your smart home gateways, they're all getting a dose of connectivity and intelligence in order to interconnect and to drive advanced features. All of those are in turn driving a richer DRAM footprint and in many cases, a larger storage application as well. We're serving those markets as well. Again, the focus in this space is a portfolio breadth that we have. If you think about what we're offering from very low density serial NOR products, all the way through very high density embedded eMMC product line.
We have a complete breadth of NOR and NAND products, a complete breadth of both legacy and leading-edge DRAM products. Including in this segment as well, we see adoption here of LP product, much like Tom was mentioning, I think it's, again, driven by the power requirements and the bandwidth advantage of the LP architecture. In this space, we're also doing quite a bit in essentially custom solutions. Multi-chip modules, in many cases, are mixed memory multi-chip modules, so a DRAM NAND stacks, NAND and LP stacks, and those are going into these applications that are sort of form factor driven, like in the wearable space or the action camera space and some of this home automation and so on.
The need to hit to a much smaller form factor is really a key part of what drives that, and to a lesser extent, some of the performance advantages of those architectures. Finally, I talked about these labs that we've built out. We have a number of them built out. You can see today San Jose, Boise, Munich, Tokyo, and Shanghai, a few other places. The focus on that is sort of a much deeper partnership with our customers, and the sort of influencers around our customers. You can see some of them on this chart. We spend a fair amount of time today with the, for example, in automotive, directly with the car OEMs, in the set-top box business directly with the carriers.
We're working with the various sort of industry consortia that are trying to establish requirements and standards in that space, both influencing what they're doing and also building upon what we're doing. Of course, we have to partner with all of the core chipset logic providers and SoC providers to help them understand the technology transitions, and sort of jointly work on what's the right solution going forward. The idea behind all of that is to enable our customers, on the right-hand side, to be able to adopt the technology, to be able to take advantage of the technology trends that we're driving, and to bring those to market much faster than they would otherwise.
This lab infrastructure has really been a strong way that we've been able to partner with those guys, help them solve their memory problem, help them solve their system problem, as Tom was talking about, looking at it from the system side, and to get to market faster with a more compelling solution. I think with that, I will move to Q&A. Everybody wants their break.
Jeff, thanks for the presentation. Can you just give us a rough, I think we have a good understanding of memory content growth per application in the various segments. Maybe if you can just give us a rough sense on the size of your different sub-segments?
Yeah. For us, the consumer, in particular, let's combine consumer and connected home so you guys can do the math easier. Consumer and connected home is by far the largest portion of the TAM. That's the largest market that we serve, followed by industrial, followed by automotive. In rough numbers, automotive is a billion-dollar TAM. Industrial is $2 billion-$3 billion, and the consumer and connected is another $8 billion-$9 billion, in round numbers. Our business is weighted actually more toward the automotive and the industrial side, where we can create and deliver better value, where the stickiness and the longevity is appreciated and valued more. In the consumer space, it's more back to the mix optimization point that was raised earlier.
Operationally, we look at that and how much supply we want to put into that on any given quarter based on the other opportunities that we have in the other BUs. Yep.
Hey, Jeff, it's Errol Hagy from Nomura. I see you guys have operating margin there of 22%, I guess, how do we view the cost structure relative to the other groups, seeing as you're addressing solutions with the broader portfolio, including SLC, NAND, and NOR?
Cost structure, we have sort of a common set of cost structure across the organization. The product mix is very different in different organizations. Part of our mission, as part of Micron portfolio is really, as you said, that 20% or 22% operating margin is sort of the target to continue delivering that level of performance as we bring new technologies in, and as we continue to deliver on technologies that we've invested in long before. Return on capital, in essence, to continue profitably driving that and leveraging that capital in long term. As Scott was talking about the planar technology, and how much and how fast we convert that, clearly for many embedded applications, we're going to continue to drive planar NAND delivery for a long time to come.
We intend to kind of leverage that capacity investment that we make today long into the future, which helps the cost structure and the operating margin.
In one of your early slides you put up, it showed the rate of growth of NAND and DRAM in the individual markets.
Yeah.
In every case, I think except the connected home, the DRAM was growing more quickly. Do you see that changing at some point in the future, just across the board for some underlying reason, or is the question really not even well-placed?
No, it's a good question. I think that there are several applications today that are driving faster DRAM growth that seems like it won't be true in the long run, right? Because I think as we think about the application space, you're going to need a certain performance, certain bandwidth. Then like your cell phone, like various other applications, it's more likely going to be the NAND footprint that's going to grow faster after that. Yeah. I think that it's more a function of the specific applications that we have and specific transformations that are happening in those applications today, I think, that's driving that. You're right. Today, our DRAM is slightly above or above across the line from a growth perspective, right? I think as longer term, we would expect the NAND portion of that to be higher. Yeah.
Jeff, thanks for the presentation. Just a quick question. In the auto and the industrial space, you saw Spansion buy the microcontroller business from Fujitsu-
Yeah
Which then got bought by Cypress. You see companies like Microchip making more investment into embedded memory. Can you help me understand from a competitive perspective where the integrated market sort of stops, the discrete market starts, and kind of how do you compete as more and more microcontroller companies are trying to beef up their own memory technology?
Yeah. There's probably many different ways to look at both of those acquisitions, but there's clearly a market for MCU with integrated flash, right? Or any integrated memory, DRAM, for that matter, or SRAM. Most of that integrated flash technology is kind of hitting a wall at 65 nanometer kind of generation. I think all of the guys in that market today are trying to figure out how do they get to whatever is after 65 for them. I think that's part of the rationale on some of the acquisitions, was a belief that technology is a way to open up and continue to explore in those markets. Today, you look at the density that's getting integrated, it's pretty small. It's pretty inefficient memory technology, ultimately, at the end of the day, right?
It's a pretty low-density application that's going into that space. I think there's opportunity for that for sure, right? Obviously. I think some of the Internet of Things build-out could very likely want to have a very low-function, for lack of better term, thin client in the IoT kind of space. I think that's where those applications are going to go. When we look at it from our perspective, we have a much more compelling NAND architecture and NAND technology advantage. We have a much more compelling DRAM and DRAM technology advantage. We have very good NOR technology that's now ramping heavily on our 45 nanometer, 300-millimeter facility. We look at that. It's interesting, but it's clearly not our only path.
I think if it were our only path, we'd go try to figure out what we need to go do about becoming an integrated MCU guy. I think we have so many other opportunities higher up the stack at higher function devices and higher feature devices, which is going to drive a better return for us.
Thanks, Jeff.
Okay. We're done? Thanks.
Okay, we're going to take about a 15-minute break, so we'll be back here at 10:00 A.M. Mountain Time. Thanks.
One, two, three, get it.
If we can get everybody to come in and take their seats, that'd be great, and we can get the day progressing here. Darren Thomas from our Storage Business Unit will be up next.
While everybody is taking their seat, I have a quick poll I'd like to ask. Just coming back from our break, maybe do something a little different. How many people, and I'm hoping I can actually see beyond the lights, how many people are using a laptop or a desktop, either in their business or in their home, that still has a spinning drive in it? Can I see by a show of hands? Yeah. The industry average is about a little bit more than 50%, and that's still a high number, but it's transforming very quickly. You're going to see that.
The good news is, I don't think I saw any Micron hands go up, I would expect them not to be doing that. What I do have here is our brand new MX200 250 GB drive. This is out of our performance line, so this is the Crucial performance line. I'm going to talk about the two swim lanes here in a minute. This particular drive, we just launched it. It's a very unique drive. It has dynamic write caching in it, which means that any time it writes, if it's not real busy, it will go to a very fast writing mode. It goes to a very unique mode, and it writes much faster than the average write speed of an SSD. It also has encryption in it like an enterprise drive.
I also have enough for everybody in this room to have one. When you leave the room today, I'd ask you to stop at the table with our support staff outside, and they'll give you one of these drives. The drive comes with the tray necessary to mount it, and it also comes with the software and the tools. The software is downloadable, and there's a key in here to use the software so that you can upgrade this yourself. If you don't feel comfortable doing that, you can call our Crucial team, and the award-winning service department there will walk you through it. You can upgrade yourself. What I will warn you is you can no longer boot your drive and go get a cup of coffee and come back. It will be done by the time you get back.
Do I have the slide? Who's got the projector advancer? Okay. While I'm getting that, let me just dive into the presentation and start by saying that the SBU is going through a transformation. We are definitely in a transformational time in our business. I've been here 11 months, as we talked about last night, and it's been my role to do a lot of transformation. What I'm going to try to encourage you to see through this presentation is that, yes, it's product. We have to come out with products like this. We have a brand-new M600 that is in more quals at one time than any of our previous drives were ever in in their entire life. It's those kind of transformations that are important. It's not just about product, it's about people.
It's about having the right kind of people, because now we're talking directly to customers, and we need people like Tom mentioned Steve Pawlowski. Somebody who can talk to the customer at their architectural level, not just about our NAND, but about their operating system, the problems they see, the headaches they have. That's what's important. It's about people, and it's also about partnerships. I'll get to a slide in about five slides on our partnerships. In the enterprise, nobody does it alone. You need the full breadth and depth of your ability to use your enterprise partners to help you execute and deliver. That helps you with time to market, it helps you with cost. It does a lot of valuable things, and that results in performance. It's products, people, partnerships, it ends up in performance.
I'm going to encourage you to see that and try to point that out throughout the slide. This first slide I have up is really about how my market really looks from the outside world. As the enterprise and the OEM customers, when they look at us in the SSD space, this is how they look at us. There's a set of enterprise class customers that carry about endurance, performance, reliability, and on time. The drive is never off. There's no idle time in the drive. That's a unique set of characteristics the enterprise industry shares. For that, they're willing to pay more for it. They use some different interfaces that are highly technical and highly capable, and that business has pretty good margins. You see the CAGR, 53%. Matter of fact, all the CAGRs of the first three are in the mid-50s.
Those are bit CAGRs, by the way. You go to the data center. Data center, a lot of people think it's eight customers, is eight data centers. The Googles, the Alibabas, the Amazons. That's the data center leaders. Those are the lighthouse guys at the data center, but almost every IT department covets that business model. They covet being able to make a data center inside their own chain link fence that operates with that kind of scale and that kind of efficiency. When we say data center, we're not just talking about the eight big data center guys. We're talking about all enterprise customers who are trying to build or are building a data center class application, and they want to buy the products that meet that.
The requirements there are they use open platforms, which means that doesn't require any specific company's hardware. They don't mind buying directly from Micron. Matter of fact, that's a preferred model. One of the big trends in this market is what we call lights out. That's the term they use. By that, they mean when they put a data center together to get this scale and cost they're looking for, they will quite often look to make the data center be so no human goes in there. There's no human intervention. Their design is typically six months to a year before any human walks back in that building again. What that means is they have to plan ahead for growth and failure mechanisms and all that. Well, one of the ways they plan for failure mechanisms is to remove everything mechanical.
No fans, no spinning drives. It's a very good thing for us because we don't sell spinning drives. The data center class people are looking for those high degrees of economics that are achieved at high scale. Very interesting. Look at the CAGR. It's the biggest one up there. A great opportunity for companies like Micron that can sell direct. The next one is client. Client is the largest of these businesses by far, and it is also probably the most competitive market we play in. The CAGR is pretty big. What they're looking for is thin. They look for the lightweight. A matter of fact, when you pick this box up, you might want to look inside, it feels empty. It's pretty thin, it's pretty lightweight. The other thing they're looking for is this low power, long battery life.
That's the dream of the laptop, the notebook user is, it doesn't weigh very much, the battery life lasts a long time, everybody likes the little Gucci thin ones. That's an important piece. The most significant part of this trend is for us, we're the business class users, what's called the commercial user in the industry. We're near the tipping point. When we say the tipping point, this is where there's a strong drive to use our technology, but it's only at the highest end, then there's a lot of spinning drives sold at the lower end. When you hit the tipping point, they just sweep their product line. They say, "From now on, everything in this product line is all spinning drives." That's what we mean when we say the tipping point. It's not parity price.
The value of an SSD is significant. The price parity is not at price parity. It's more like 2.5x, maybe. That's the number everybody throws around in the industry. As I noticed on the slide here, we're getting near that tipping point for these client drives. This is pretty much an agreed to in the industry comment. You'd get the same thing if you asked the OEMs. Then the consumer space. It's probably the most different of these. Consumers about people doing upgrades. If you're smart, you'll get one of those SSDs and you'll put it in.
You'll do an upgrade, or you can get the technical brother-in-law or family member, or you can get our team to help you walk it through it, but you've done an upgrade and you're going to turn the laptop that takes 1 minute and 30 seconds to boot into one that takes about 18 seconds to boot. My joke is used to when I loaded one of my programs, it named every person who ever wrote the code. It named all the writers of the code, and you could read their names as it's loading. You can't read their names anymore. That's one of the advantages. That's the consumer market in the SSD space, and it's a lower CAGR , mostly because as the new notebooks come out with an SSD, there's less room to upgrade. You can imagine that market would start to slow down.
I want to talk a little bit about the trends in the industry, the key is, I'm going to go around this from the upper left-hand corner around in a circle here. The enterprise space, if you look, there's three different interfaces. I want to point out that we talk about interfaces because in the industry, those of us that live in this industry, the interface means a behavior, it means a workload to us. It's not about the interface. When somebody says SATA or somebody says PCIe to somebody like me, when they say SATA, they're saying a server-based storage device that's used by low-cost servers. That's what that means to me. When they say PCIe, they're talking about a very high-performing workload, something like oil and gas or genomics or military application or something like that, very high speed, high-speed database, something like database acceleration.
When you say SAS, you're talking about external storage. The interfaces, we call them by the interfaces, but they have very different workload behaviors behind them. What I want to point out here is that the PCIe interface in enterprise is growing. It's actually the fastest growing. You can't tell it from that chart, but it grows at 100% year-over-year through 2018. It gets up to about 36% of the total. The more interesting one is SAS. SAS was always a bigger technology. As the external storage companies are beginning to adopt and deploy SSDs in their applications, you're seeing the SAS market grow. It's growing at 72%, but by the end of this 2018, it's 45% of the market. That's why SAS is so important to us.
This is a fast-growing industry from a big base, there are lots of SAS drive applications out there. Coming around to data center, there's really three there, but if you notice SAS at the bottom, it's very small. Data center folks really are after a more cost-effective deal. They're typically not the highest performance model. You see that very, very thin blue line at the bottom. Most of it's PCIe and SATA. What you're seeing here is the data center guys are still very strong in SATA because it's a very cost-effective design for them, and they put a lot of the redundancy and reliability up in the application layer, so they can use what is almost a client class drive. The difference here is it's always on. The client class drives have an 8-hour duty cycle.
It's 24 by 365, 8 hours a day duty cycle. The drives are beefed up a bit. Basically, SATA drive meets their requirements, you're seeing some PCIe coming in, but this is not the PCIe from enterprise. This PCIe is the low-cost version. The server processor will come with PCIe interfaces and enough of them in the next generation of processors that you'll begin to see architects putting the PCIe drive directly on that PCIe bus of the processor. What you're seeing here is really a cost savings move to go to PCIe. That same thing occurs in client. The client drives, they're not looking for that screaming performance the enterprise guy does in PCIe. What they're looking for is that cost savings in PCIe. You notice it changes.
What I want to point out is it looks like it's a big number in 2015, but that's mostly one supplier. It's mostly one vendor out there. By 2016, Intel will have processor chips that have lots of PCIe interfaces, we can put drives directly on them. By 2016, you'll see the rank and file OEMs starting to move to the PCIe bus. It's going to be a transition. It's not going to happen overnight. This is not something that the big notebook guys are going to just redo their whole line just to do this. As they update their lines, they're going to go to PCIe. You see the consumer side. PCIe kind of follows. The consumer side is really the same as the client side, just lagging a little bit because it is the upgrade market.
The trends roughly here are, what I would walk away from this is, SATA is still big, as we walk through this. SAS is enormous in the enterprise space, and PCIe is coming, and is coming on strong. That's the walk-away message from this slide. Now I wanted to talk technology, and I'm going to limit myself to referring to these by products. If you want to ask a lot of technology questions, I'm going to have them re-mic Scott up, because this is really his area. But what I want to point to is from a go-to-market standpoint, and this is looking at SSDs. If you look at the light blue space on the left, what you'll see is, before 2015, and I'll make a point, this is the transitions as Micron is doing them.
If I haven't put TLC as soon as you think it hit the market, that's because we are coming out with it later this year, this is where I showed it. If you look on the left-hand side, you'll see that MLC met every piece of the market. It met this consumer retail market. It met the client in its day. It was priced effectively. You'll see the data center drives still need the MLC for the endurance and capability. You notice it met the enterprise space. It was a universal technology that fit across the product set. I think some of the questions the team was asking Scott was about, why is the change in TLC and planar? What I've shown is the light blue space in the middle.
This is planar TLC, and this is looking at it from the way a market person looks at it. What you'll notice is planar TLC is dominating in the retail space. Obviously, it's driving the prices and it's affecting every company, including us. If you look at it in the client space, you'll see the same thing. Planar TLC is a good product. It's good enough. From the description I just gave you on the previous slides, the client space doesn't have the same performance requirements. They don't have the same endurance requirements. None of the requirements are the same. TLC works fine there, but if you'll notice, we have no intention on moving them up into data center. They might make it into the low end of data center, maybe a less than one DWPD per drive, because the data center sometimes buys client drives.
For the true data center, the one that's always on, never off, lights out, doing what we talked about, TLC, the 16 nanometer and below TLC just isn't good enough. We can't reach the DWPD, the endurance the customer wants, or the performance they want with that drive. Of course, the enterprise is a step function above that, and it certainly doesn't meet it. What I'm showing you here is, in a way to think of Micron's desire and insight years ago was that 3D TLC, if you'll notice, 3D TLC meets it all again. That's the technology that, at least in my SSD space, gives me the product I need to be competitive in the market across all of these technology. Gives me the price. He showed you how it's a better price than 16 nanometer.
It's way better performance and way better endurance. Significantly better. The 3D is the Micron future for SSD. It's the industry future for SSD. I would leave it to Scott to explain to you all the technology, but I'm just showing you that this transition has followed this pattern, and it will continue to follow this pattern. The technology deployment for us is we will have a planar TLC in our mid-range and the low-end technologies. As we go up, we can't deploy them there. We can deploy the 3D TLC very cost effectively with the performance that the customers need. Now, just walking through this, I want to walk you through a little bit of this transformation I was talking about when I opened the slide. No excuses, we are a product company.
We have to have a SSD portfolio. We have to have technology. I think Scott's done a great job of showing you where our technology's coming, and some of this is you just got to sit back and watch these things happen. We have to have a great portfolio. Now I want to talk about expanding the enterprise capabilities, because the portfolio is absolutely on its mend. We're improving it. This drive, the M600, a lot of technology. The announcement we did last night. We got next generation drives coming all the rest of this year. There's lots of product here. To make that product really reach our end user customer, we have to change a few things about us. First of all, we've hired an enterprise, or we've expanded our enterprise class sales team.
We hired a gentleman. Mark Adams showed you his name, Mark Glasgow. Mark Glasgow comes from the enterprise sales side. He's a true enterprise sales person who is used to talking to 25, 50 customers a week. I'm talking end user customers. This is not negotiating with one of our OEMs. This is actually learning from the end user customer what they want, what they need, explaining to them our portfolio so that they can buy our technology either directly from us or through our OEM partners. We just really expanded that team. The team had a big kickoff earlier this week. That team is now up and running. We also hired, like Tom Eby hired Steve Pawlowski, you might have noticed an announcement from us. We hired a gentleman named Rob Pegler.
Rob Pegler was a CTO at EMC. He's now on board with us, also started earlier this week. Rob is one of those people that can sit in front of a customer and talk about their software, their applications, their OSs, their networking, all the way down to anything they want to talk about, all the way down to how the drive can make a difference in those applications. As you might note, when you talk about this class of people, Steve Pawlowski and Rob Pegler are kind of in the same class. They're birds of a feather. These two gentlemen will spend a lot of time with each other. We're adding to that capability. That's another example of the enterprise capability. We've also built qual teams. I told you we're in more quals now than we've ever been in our history.
We put qual teams in countries where we sell these products. At one time, most of our qual teams were here in the U.S. We now have qual teams overseas where the customers buy these products, where they deploy these products. We've expanded that. We've expanded our firmware team, and we've expanded even my marketing team to include people that know how to talk to end user customers and have that access to that process. Expanding the capabilities. Strategic partnerships. I think that I'm going to mention the Seagate relationship. We talked about it a little bit last night, but I've got a slide following this where I'll roll it out for you again. That's not the only strategic partnership. We've had one with Intel for a long time, and it's not the only one we're going to do.
I'm just going to say stay tuned. We're going to continue to do these strategic partnerships because I can do more with the enterprise class partners. Together, we can do more than we can by ourselves, and one and one is three, and we'll both make more money doing these things. Strategic partnerships are a very important part of our business. 3D NAND, I've shown you the slide. It's our future. When we get to 3D NAND, we will have the product we need, we'll have the technology we need, and it just keeps getting better after that because we have next-generation memories, which I'm not going to talk about, but just point to the fact that these things are on the horizon, and my team knows how to deploy them and use them with our customer and our customer relationships.
Finally, something people don't ask a lot of questions about is software. Now, I'm not talking about firmware, I'm talking about software. Software means our ability to influence VMware or Microsoft or Oracle or SAP or something like that. I hired a software vice president. I think I mentioned it at the Hong Kong team announcement when we did the investor meeting there. He's been busy at work. We actually opened an Austin design center. There is now, we have an office complex with over a dozen people in there writing code, communicating with our partners, making sure that if there's a way that Oracle can use our systems better, we're working on that way with them. We're partnering with them at a software level.
A very important piece of a transformation for Micron, to have this ability to make the world better and for us to dictate how it gets better. Here's the Micron Seagate announcement. As we've mentioned, it's a multi-year agreement. The way you should think about that is the enterprise customers want to know that we're not speed dating. This is a real deal, it is a real deal. It combines the innovation and technical expertise of the two industry giants. Seagate's got some excellent technology, not the least of which is their SAS technology, which they invented, and now we have access to that. Micron has not only 16-nanometer NAND. Our first product coming out will be a 16-nanometer NAND product. We're going to have an enterprise class 16-nanometer product. That's the kind of innovation that we can do together.
The initial collaboration is the SAS drive, but it's not limited to that. That was just the one I was the most excited to have. Seagate has a lot of other technologies. They have flash arrays or all-flash arrays. They have many other technologies that we might be interested in, and we have technologies they will be interested in. It gives us, obviously, we have this access to their drive and their other technologies, but they have access to our strategic supply. For them, what we're doing is we're bringing them in the fold and saying, "Look, we're going to make sure that if you forecast and manage your inputs to us properly, we will make sure you get your volumes." Finally, it establishes a framework for future collaboration.
This is a real enterprise class partnership, and it will result in real measurable inputs and real measurable results for Micron. I'm not going to go into any more detail than that, but to tell you that this is what we've been missing. I wasn't playing in that big blue space. That was one of the most lucrative SSD places. In a very short period of time, I'm going to be playing there. That's very important. In closing, my intent is I needed to build a world-class team, and together with Tom's team and my team and Mike Rayfield's team, and we're combining the best of the best together, and we're building and hiring these resources, and we're sharing them.
Just so you know, Stephen Pawlowski, I use him probably half as much as Tom does, but my team uses him because he's an expert at these kind of things, and likewise with Rob Peglar. The world-class team's important. Innovation, we're going to continue innovating, and we're going to start taking advantage of partnering and partner innovation, and then engaging. We're going to engage with end user customers. It's an important part of our transformation, and it's what will make Micron be seen as a technology leader, not just by our OEMs and the people who buy the technology from us today, but by the end users, the people who use that technology, and that will redefine the future. Q&A. Let's see. I think I beat the microphone people.
Hi. Thanks for the presentation. Maybe could you talk about what is your enterprise SSD revenue right now? What do you target for over the next two, three years?
Yeah. What I don't want to do is get into a prediction, because the revenues are going to be driven by a lot of factors other than just me doing this. Do we state in public?
We don't break out our enterprise revenue specifically. What we've said is SSD overall is about 20%-25% of our trade NAND revenue overall, and I think on our earnings call, we said roughly 25%-30% of that is enterprise today.
Yeah. By the way, that's not enough, right? That's the key. That's not enough, and we want to raise that because the larger the enterprise revenue, the larger the margins will be. In the back.
You mentioned last night that both you and Seagate will be able to take this SAS SSD to market to your individual customers. Are you able to start selling this SAS SSD that they already have immediately, or you're going to develop a SAS SSD and it's going to take some time to get that qualified in?
Yeah, it's a great question, I actually asked that question, and I'm not sure anybody would've allowed me to sell that drive before. No, it turns out that the drive had a very limited volume to it, and I couldn't get any volume. No, we're not going to sell the current drive. We are going to wait and sell the new drive, which is going to come out in a very short period of time. Let's see.
Hi, thanks. Mark Newman from Bernstein. As you talk about building out your capabilities, and hiring people in the enterprise side, are you thinking about actually building out an enterprise sales force and selling directly to enterprise customers, or do you think your main channel is going to be through the incumbent OEMs, EMC, et cetera? Then a follow-on question related to that is, could you comment on between enterprise and data center, the profitability in those segments?
Okay. Let me take the first one first. It's a great question, and I'm glad you asked because I didn't want to leave you thinking that we're going to go hire 15,000 salespeople and try to be in every market and every place. That's not the model. The model is really very small, what in the industry is referred to as sell with teams. We would go to customers, and we would partner with somebody like a Dell or an HP, and they might bring us a deal and say, "Micron, you might want to talk to this customer." It'll be a very focused and targeted approach with our OEM partners, and they will help guide us on that. We may find some deals and bring it to them. That is not going to be a sweeping across the world or nation kind of deal.
The team will focus on the data center class accounts. The team does have the responsibility to focus on the data center, much smaller group of people. I would expect this team to be relatively small. It's not going to be the size of our regular sales team, but it's going to be very focused on selling enterprise class. You go sell to data acceleration guys, or you go sell to oil and gas guys. You might hit all 10 of the oil and gas companies, or maybe five or six go sell with some Oracle partners or something like that. It'll be very targeted. The second question. I had it in my head a second ago.
Difference in profitability.
Oh, profitability. Yeah. I guess I could say there is a difference in profitability. It's still better than client, but it's not as good as enterprise because those customers are willing to accept a less robust drive. The profitability is somewhere between the two, is probably the best way to say it.
Hi, Karl Ackerman from Cowen and Company. Just curious, what are your thoughts on some of the larger hyperscale customers buying raw NAND, making their own drive? I guess, do you see this altering the competitive balance for drive suppliers, and what's maybe your thoughts on if others pursue this integrated hardware strategy on your partnership over time?
I'm not sure I heard the last question clearly. It was about the partnership, if other competitors
Other hyperscale customers, as they adopt this new enterprise SSD product.
Okay. I'm going to take the last one first. Well, make sure I understand. You're asking if the enterprise customers are going to notice the difference between the two drives and will care?
No. As some of the hyperscale customers, if they adopt more towards just buying raw NAND-
Yeah.
How does that impact?
First of all, the hyperscale customers, there are a group of them, a very small number, that can actually build their own SSDs. They have enormous engineering teams, and they want to buy NAND. We're happy to sell them NAND. That group of people is, first of all, it's a small number. Not everybody can build their own SSD. It's not a super simple task. There are advantages to doing it yourself. You can change form factors, you can do all kinds of different things like that. We're certainly in support of that, and there will be a set of data center class, if you will, the top maybe four or five that can do that. Most companies don't have that capability. They just don't have the ability to do that integration. They will buy from us.
The enterprise relationship that we have with Seagate now will allow us to sell in any one of those categories. If somebody wants to buy the drive from us, including data center, or we can sell the chips. The relationship does not cover us breaking the drive apart and selling the parts.
Hey, Darren, it's Errol Hagy from Nomura. I guess, how do we look at the moving pieces related to software, firmware, and controller technology as a result of the Seagate partnership, and what's your strategy there?
The hardware and software, say it again?
It's software, firmware, and just the controller technology.
Yeah. We actually didn't break it apart by components. What I'm really doing is I have a relationship with them to buy the same drive coming off the ODM factory that they have. They will still manage the software, the firmware, the controller. My team will buy those same parts and build the same drive. In a way, they're my engineering team. They're the team that's responsible for making sure that the components all work together, and that's all part of the relationship. What we're doing is just really acquiring the drive like you could just go buy it from an ODM manufacturer. It's much simpler than me having to go buy individual parts and then try to manage firmware, software, hardware. The intent is that the products are identical. Now, this is key.
The products are identical so that when I go to an OEM, if they want to get two drives for the price of one qual , they can do that because the drives are identical. Other than the fact our name is written in ours and has a different label on it, their name is written in theirs, has a different label on it, the drives are identical. Electrically, they're identical, and firmware-wise, they're identical.
Darren, we've got one more question over here, okay?
Hi, Darren.
Hi.
What's your view on using an SSD that talks directly to the memory bus on a CPU for higher performance? What size market do you think that is?
Yeah, it's interesting. That market's kind of got two flavors to it. One is where that, and you're talking about the DIMM sockets, where you put them on a DIMM, and there's people who put them on a DIMM and talk to it in a memory semantic, so it still looks like DRAM. That technology is interesting to us. The CNBU team is actually looking at those products, and that's actually pretty interesting. It actually ends up being like a power-saving version of a DRAM or a power loss protected version of DRAM. The other one is where you actually change the semantics. You now actually, you put it in a DRAM socket, but it looks like a SATA drive. There's a few industries where that's interesting. It's really the older industries where they've used older software and been trapped by the older software.
If you do that, the system gets a little bit faster, and it gets a lot faster. It's helpful. The problem is there's very few DIMM sockets, and customers are still actually putting DRAM in those DIMM sockets, so it's not like if they have 36 DIMM sockets, I have 36 hard drives. The customer's going to use 20 of those for DRAM, so I have 16. 16 in these cards are very small. The problem is it's just not a scalable answer for many customers. It's good for immediate, almost like somewhere between DRAM and storage, like for mini storage acceleration. It's good for that, but it's very limited in scope to people who can fit in that size capacity. The market hasn't shown itself to be very big. There's been a few startup companies that have tried it.
We're certainly supportive of them, when they do it, we talk to them as part of my partnering responsibility. We would absolutely look at them, if the market gets big, we'd certainly want to participate. Right now, I think the market is still TBD. I've got to see the market more. There's been a few very interesting applications, but they're almost niche. We're looking for that market to get a little bit bigger before we commit. Is that it? Okay.
Yeah.
You're staying right under the two lights. All right. Well, thank you, I'd like to introduce Mike Rayfield, our mobile VP, I'll give you the controller, Mike.
Thank you. Thanks, Darren. I really appreciate you all participating in my twice-annual public performance review. I actually get a lot out of the conversations we have over the two days. I learn probably as much as you do. Today, we're going to talk about three things, and it's pretty consistent with the last four times that we've spoken. We're going to talk about the amazing sort of acceleration and capability of mobile devices we've seen over the last couple of years and where we think that's going, the capability both in terms of memory content and storage content. We're going to talk about a change that's happened in storage in mobile devices. If you looked at the NAND business in mobile a year ago, a year and a half ago, and forecasted it was all something called eMMC, which is basically controller in a NAND.
Well, on the sort of journey to a $2 billion smartphone business, people wanted to get to production much quicker, have a simpler assembly process, and what was born was high-complexity EMCPs. The reason that's important, and we'll go into more details on it, but it requires NAND, DRAM, firmware, packaging technology, and controller technology, and there's very few people that have all those. That's been a huge opportunity for us. The last one is, we've talked a lot about, the last couple of years, about how do we go to market. How do you go to market and be successful in a market where it's a very concentrated customer base? All along, we've talked about it's actually becoming less concentrated. There's innovation happening in areas where it isn't as concentrated and sort of the rise of what used to be China, Inc.
is now a bunch of named customers that are really doing some pretty significant innovation. Let's start. If you'd looked at this slide a year and a half ago, it would have actually had some different categories. We put high-end and mid-range smartphones together, and the reason we do it is, quite frankly, the whole mid-range of the smartphone market is now exceeding the high end of the smartphone market in terms of growth, size, and functionality. Memory content in the mid-range of the market is significantly higher than the high end in both DRAM now and it ultimately will be in NAND. If you look, the next box, not that long ago, would have been feature phones.
The reality with things like Android One, we've got smartphones at under $100 that have a gigabit of DRAM that have fundamentally started to completely replace feature phones, and feature phones had very little memory, as you know. That's ultimately what drove 550 million smartphone units being sold in the fourth quarter of last year, is because you could buy so much for so little, and the differentiation a lot of times was in memory content. Tablets, architecturally exactly the same as smartphones. As we ended up with six-inch displays on smartphones, it's hard to tell a six-inch smartphone from a seven-inch tablet. I think going forward, we're going to start to see those things merge. Again, architecturally the same. They're going to care about the same things. They're going to have the same applications.
They're going to need the same experience, which historically was better on a tablet, so it's going to force a better experience on the phone. I think that'll continue to be a great driving force. The last thing is smartwatches. It's sort of been parsed out of what this Internet of Things was. The reality is it's now adopted the smartphone architecture. We'll talk a little later, but the amount of memory, the performance necessary in that device, has turned pretty significant. It's going to be hubbed with your tablet or your phone, but the things you expected out of that are far beyond what I think people would have even thought a little while ago, and that's driving significant memory content as well. Let's see if we can quantify a little of this.
When we talked in August of 2013, we had this conference in New York, I mentioned this little company that almost nobody had heard of, this little company called Xiaomi, and they had this phone called a Redmi, and nobody had really known much about it. It was about a $200 phone. It had 2 gigabytes of DRAM. They had a 3 gigabyte option coming, and it was a couple of hundred bucks. I said, "This is what's going to change what happens in our industry, because it's a really amazing product." The last two years, I think everybody's heard of the company. Ultimately, there's a bunch more companies like that driving the content. If you look here, in 2014, high-end phones, which include that mid-range, is about 1 billion units.
Again, the units that are in the mid-range, the $300 phones, are the highest functionality. My main phone is a Chinese phone. It's $349. It's got 64 GB of NAND and 3 GB of DRAM. It's an amazing device. This is the kind of stuff that's getting put out and driving the numbers we're seeing. Fundamentally, we thought asymptoting at maybe 4 GB, but the reality is we saw phones introduced just this year at CES that had 4 GB of DRAM in it. If you start thinking about the amount of memory in your computer at home and the significantly higher performance task you're asking your phone to do, you can see that continue to grow. If you look at tablets, it sits almost on top of smartphones, as we talked about.
Again, because you're going to want the same kind of experience, we're going to see those in the neighborhood of 4 GB pretty quickly, and we're already seeing some. Asus, Xiaomi, a bunch of folks have announced 4 GB phones already with both LP3 and LP4. Then smartwatches, I think if I had quizzed people before this and said, "How much memory DRAM is going to be in a smartwatch in a couple of years?" Most people would have come up with 256, 512. The reality, it's going to be 1 GB moving to 2 GB. A lot of it is you may capture some video, you may want to display some video, you may want to compress some things, you may want frame buffers, but that's going to drive a couple of things.
It's going to drive packaging technology, so it's going to be EMCP. You're going to need NAND, you're going to need DRAM, you're going to need controllers, you're going to need firmware, you're going to need packaging. It's going to drive staggering numbers. This is going to be, I think we see, a growth driver that many people hadn't thought through in the past. Let's talk about this shift in managed memory in NAND. Again, two and a half years ago, when we sort of started on this journey of building up our mobile business, almost all of the NAND in phones was eMMC, and we didn't have a very good position in it. We started investing aggressively in firmware. We've invested aggressively working with partners on controllers. There were five people or six people that we competed with.
In this shift to try to get phones to market more rapidly, we have seen all of a sudden significantly higher density EMCPs. As you'll see from the chart here, it used to be 2 GB and 4 GB. We're seeing configurations now that go up to 32 GB of NAND and 24 Gb of DRAM on an EMCP. If you've got NAND, if you've got DRAM, if you've got controllers, firmware, and packaging, the opportunity is significant. I honestly believe this talks about 60% of the market. The opportunity exists to be greater than that. It's put us in an outstanding position. You combine that with sort of the focus on many of the emerging markets and the emerging customers where they drive this, it's been a great opportunity for us.
You've seen the data on our conference calls about how rapidly the mobile NAND business is growing. This is what's driving it. We've talked about the markets and why we're interested about them, but in the end, it's about solutions, right? How do we differentiate? LP3 and LP4, making sure we've ramped 25 nanometer, we've qualified it in the majority of our customers. As Scott talked about, we're ramping 20 nanometer. That allows us to build 6 gig and 8 gig LP3 and LP4 parts, and why that's more important now than it ever was, it allows you to build a 3 gig phone and a 4 gig phone on a four-stack part.
Working aggressively to make sure we've got what we think will be the best technology in there and the best packaging to be able to go off and serve that upper end. All the firmware that we've invested in, the team that we've grown, there's two places where we use that firmware. You use it in eMMC. When we talk about that high-end, which was supposed to be the whole market, but it's the high-end now, we still service that. We can go off and partner with our high-end customers to sell that.
The 60% of the market and the one that is the most exciting to me is that same firmware and controller work goes into the EMCPs, and we've got LP2, LP3, LP4 EMCPs that you will see in phones that are ranging anywhere from $70-$350, and the functionality of what's now an $800 phone. That's the shift, I think, that all of us thought might come. I think it's come significantly faster than we'd anticipated, and it plays well for Micron. Watches, we talk about this, it's about power, it's about energy efficiency, it's about a small package. Again, people are going to want to put as much as possible into that package. What we'll end up with is there'll be a small processor.
On top of that processor will be what's called an ePoP, which is basically an EMCP mounted on the top of the processor, and that's all that's going to be in the device. There'll be a PMIC and a radio off to the side. We think that all the work we've done in the EMCPs, allowing us to learn a lot about ePoP, allowing us to learn a lot about packaging, and the work that Scott's done, is going to put us in a great position as this goes off and grows. This is going to grow with the big phone guys. Ultimately, you're going to tie these things to your device. You're going to find a watch you like, and it's going to make you like your phone even more, your tablet even more.
They're going to go together, and we think it's going to be a pretty amazing accessory market, if you will, that'll drive significant DRAM content. Historically, people thought it'd be relatively low. I think both DRAM and NAND content will be pretty exciting. When we talk about go to market, I think when I started speaking with you a couple of years ago, it was all about the high-end isn't growing anymore. There's only two customers in the world, and that's going to be a challenge for you, Mike. You're not going to be able to figure that out. The reality is, where innovation's coming from now is not only the big guys, they build amazing devices, but innovation and functionality and price performance out of China. The largest cell phone supplier in India, most of you had never heard of until very recently. Micromax.
Working with people like that, working with the people that do the chipsets for those folks, all those small guys down in the corner, there's Indonesia's largest handset guy, Brazil's largest handset guy, India's largest handset guy. Finding a way to work with those folks and come up with solutions that allow them to get to market very rapidly will ultimately become a larger and larger percentage of our business. A year ago, when we talked about China, it was China Inc. Now there's six or eight named people you all know, many of them approaching 5% of the market. Having relationships with those folks, doing products for them, and ultimately having them drive this eMMC business of ours has been a great opportunity for us, and I think how we looked at the market made a big difference in that.
Finally, I think while this is a 2 billion unit a year opportunity, I honestly believe, and the thing that excites me most about it is mobile computing is sort of just starting. It's the first market in the history of the world, I think, that you can call the 2 billion unit opportunity to start. But if you think about tightly coupled SoC architectures with a combination of NAND and DRAM together, either mounted on top of them or very close to them, that architecture is going to go to all these adjacent markets. I think all the learning we do here is going to continue to go further and further.
I think as you get flexible displays, you get wanting a greater level of functionality, but no compromise in the experience on things around your house, things around your wrist, things you wear, it's going to leverage this development we've done, and again, probably dwarf the current market. I think that's what's so exciting about this is we're just starting, and it's already pretty big and exciting. With that, any questions?
Thanks for the presentation, Mike. This is Harlan Sur from JPMorgan. Good to see you moving up the value chain, looking at eMMC, EMCP. One of the issues, obviously, with a module-based solution is you're purchasing a merchant controller solution. You've got the overhead of the module manufacturing. Help us understand how Micron is sort of optimizing that overhead to continue to drive pretty good gross margin profile in those module-based products.
Great question. I look at the list of things where I can be most impactful. Building the right NAND and right DRAM, obviously. Differentiation through firmware is critical. Then as my business grows, I'm a pretty significant partner to, it's been announced, Phison is my controller supplier now. I'm significant enough that I can work with them and say, "Hey, I'd like you to make this. This is what I want." I get the leverage of basically a semi-custom controller. They have the team to go off and do it, and I can focus on things where I think it's a higher order bit in terms of accomplishing things, and that's firmware and my own base technology and then packaging. Out in time, will I do some of my own controllers where I think I can differentiate? Sure.
Right now, over the last two years, it's been, let's get our footprint in the marketplace, get the design wins, make ourselves successful, and then be able to go off and differentiate in other areas later.
Thanks. Sandeep Baliga, Jefferies. Can you talk a little bit about timing of LPDDR4 as well as 3D NAND in mobile, both for Micron as well as for the industry? Thanks.
Sure. LPDDR4, we've got parts at our customers. We're internally qualified. We're ready to ship to customers when they start to ramp. The LPDDR4, DDR3 trade-off is going to be an economic one, right? I want to get the design wins. It's deciding when to ramp that hard. When it's a significant percentage of the market, which is probably next year sometime, I'll have my fair share of that, and we will have sort of traded off, to make sure economically it's the right thing to do. I'm pretty comfortable with where we are on that. I like the ability to trade off those things. LPDDR3 is going to be the workhorse for mobile for a long time. We're going to go off and make sure we've got the most cost-effective solutions there as well.
In terms of 3D, I think 3D is going to bring great things to mobile, both some MLC, but also TLC. As soon as Scott's got that ready to roll, we'll end up utilizing that. The impact of Apple introducing really high-density NAND devices is going to go everywhere. I think 3D will help us get to that. I think that you will all have 64 or 128 gig of NAND on your mobile devices pretty quickly, because it's going to be available, and you're going to want to store all your data, and 3D is going to help us get there with great performance.
Two questions. On the mobile side, do you see any risk to increased content as operating system improves? Is there anything out there that you see is slowing down the increasing of the content? Then second question, when do you see the app processor and LPDDR integrate into the single package? Is it more of next year, or is it maybe couple of more years out?
In terms of efficiencies of operating systems and things, I think the reality, if you look at what you do in DRAM, display resolutions, multiple frame buffers, textures and graphics, all of those chew up a staggering amount of DRAM. Now I'm going to add a couple of 4K frames as I play my small 4K video clip. There's really nothing that forces us to go in the other direction that I can see. I've talked to a number of you want to play with all these different phones. The minute you see a stutter as you move around from screen to screen, you almost get this visceral reaction. You don't want it. Ultimately, the phone guy's got one shot at you, and they want to make sure it's as good an immersive experience as possible.
They're going to over-provision the memory, because it's the simplest thing to make sure you get great performance. The next piece, in-package memory. I think memory's going to get closer and closer to the processor, whether it ends up sort of a modified POP or ePoP, whether it ends up with some sort of unique high-speed interfaces. I think there'll be some people that put a small amount in the package out in time. It's going to be driven by, does it really make a performance difference?
You see that more of next year or more of 2017?
I think it's beyond next year. It's beyond next year. Yeah. Ideally what you'd do is you'd repartition what the memory storage interface looks like when you do that, and there's only thought process on how that works now. It's a ways out.
Mike, thanks for the presentation. A couple of questions on applications that could drive more content of memory into phones. The first, I know it's too early, any update on 4K camera modules actually getting into the phone and people taking 4K pictures, videos on the phone itself as content creation? The second one, I think Apple just recently upped the maximum gigabyte per application allowed from two to four, in their App Store. I'm trying to figure out what does that mean for DRAM? What does that mean for NAND from your perspective, if that's true?
I think, I don't know the schedule of the 4K sensors. I do know that 4K content's being streamed now, and people are going to want to start looking at it. I was laughed out of the building seven years ago when I said 720p would be on phones, and now 4K is going to be here. Clearly larger applications help in terms of the requirement for more memory. I think the thing that is unique is I think the NAND content is really going to start to go up pretty dramatically. Now that you can get very large EMCPs. I've got a case study of one. I've got a 16-year-old daughter who's never had more than eight bytes of NAND free on her phone, no matter what it is.
Those are the people that are going to go off and buy the phones for the next 40 years, right? They don't know what's on it, they don't care what's on it, they want more of it. I really think, as people get larger and larger devices, memory-wise, DRAM-wise, the performance is going to get better. NAND-wise, they're just going to feel better about having their content.
Last question over here, Mike. Go ahead.
Thanks, Mike. Doug Freedman, RBC. When I look at the target market for you have four areas where you highlight in your slides. You've got Apple and Samsung on one side, your other markets. Can you talk about your opportunity at Apple? You're presently selling them DRAM, but I don't believe you're presently supplying any NAND to them. Is that something that you target? Then how should we think about your relationship with Samsung and what you might hope to do there as well, given that they do have their own internal offerings?
My objective is to be a great supplier to everyone in the phone business. I spend time with, I think, every phone company on the planet, and there's a lot of them. I think that as I get better and better solutions that solve their problems, that do a better job or support them, anybody better, I think I'll be a better supplier, and I'll have a better chance of doing business. I have no problem doing business with my competitors. I have no problem doing business with the largest guys and the smallest guys. In terms of what their plans are, everybody sort of architects their phones differently, does their own things, those decisions will either allow me to be a big supplier or not.
So far, I think that the market is pretty happy with the portfolio we've got, and we'll continue, I think, to make it better and better so that hopefully they all call and want to do business with us in all the different areas. Did I avoid talking about a direct customer well enough? It's hard. That it? Great. Appreciate it. Thank you. Bye, Mike. Okay. Good morning to everyone. I'm going to do a wrap here of the business units, talk a bit about the innovation opportunities across this entire spectrum. We'll go ahead, and we'll talk a bit about the model, then turn it over to Mark Durcan, as well as Mark Adams, for a final wrap and Q&A.
Hopefully, this morning, what you got a feeling for across all four of these business units, is really just the staggering potential in front of the existing product portfolio and how excited we are about the opportunities driving that, both from the demand side, the innovation side, and really, where memory and storage are adding value out into the industry today in ways that they never have. Frankly, what we feel is even more exciting than that is that behind, really, call it this first layer of innovation, there is a sea change happening to the real computing system, across the entire landscape, top to bottom. The opportunities that that opens up and how we think about those, the role that memory can play there is, frankly, I'd say, probably the most exciting thing in front of the company today.
It's a set of opportunities that I think we're just scratching the surface of. I want to talk about a few of those and what we're doing about it specifically, a bit segment by segment here. You heard Jeff Bader, Mike Rayfield talk about the Internet of Things here, obviously enabled by large-scale connectivity, getting computing moved out in ways that it never really has been before. What that means to a company like Micron is, frankly, an opportunity to move memory and storage subsystems in platforms that really have never considered themselves consumers of memory before. Frankly, this is a space that we're in, really, the first inning of, to put it bluntly. Absolutely explosive growth happening, this is true all the way from industrial connectivity, all the way to the watches, et cetera, the client applications.
There's a common thread here, which is that you have a very wide, diverse number of enablers of application processors, and increasingly, they're looking for ways to get memory and storage subsystems, high-performance subsystems, by the way, to their application. Really, the things we care about here are ease of use, and probably a performance characteristic that focuses on power and battery life like no other, given, obviously, the characteristic of the application. Really, an innovation opportunity where we're looking at the DRAM, the NAND, the NOR portfolio, even forms of emerging memory and saying, how can we combine these to get the power of the application down as well as the ease of use enabled much more quickly? Mobility. This is a biggie, and I think you got a good flavor of that from Mike's presentation. Here, you're talking about, frankly, the Internet of Things.
If you think about the Internet of Things-based platform, mobile, five, six, seven years ago, looked an awful lot like that in terms of the application processor and the memory subsystems around it. You think about it today, these systems are just as concerned about power. What they're much more concerned about now is actual performance, and that is a goldmine, quite honestly, because the keys to better performance lie in better memory systems, better storage systems. What's happening here, Mike talked a bit about LP4. The fact of the matter is, you have a prerogative here of how do you get much more DRAM and larger storage subsystems closer to the processor itself, keeping the power down, and making sure that we can deliver a much better computing subsystem at acceptable power. You get to networking.
This is a segment Tom covered, and this has been a good segment for Micron. Frankly, it represents really the first of the big segments where memory innovation was not just critical, but the only way in which new systems could advance. We're seeing that today. Some of what Tom talked about with HMC, before that, it was Micron's reduced latency DRAM, where there was no other way outside of a much more advanced memory, something that was not offered by JEDEC, something that required real memory innovation to go do. The fact of the matter is, the next generation systems cannot be accomplished any other way except for this kind of innovation.
We were talking earlier in some of the Q&A about partnerships, the kinds of companies coming to Micron looking for innovation just simply to go enable and deliver their next generation of high-end networking infrastructure, the big platforms, the control and data plane architectures necessary to get video moved through, be it the 100G standard and soon the 400G standard. It just can't be done outside of memory innovation. Cloud computing. This is the broad volume server space today, the data centers, virtualization. What we care about here, and the innovation opportunities come down to a prerogative of making sure that not just in the main memory, but also in storage, there is very low latency access to whatever the hot data is, okay? That hot data can move around all the time.
There's a prerogative to make sure the latency across every single one of these memory and storage hierarchies is minimized as much as possible. You think about the gap in between memory and hard disk drive that existed initially, a relatively small gap 20 years ago, and that widened over time. Obviously, NAND has come in to fill that gap. Here, you have the same kind of gap opening up that gives the opportunity for true storage class memory, and I'll talk more about that. What storage class memory does, what PCIe drives do, even innovation inside of the DRAM subsystem, is that you have a much better opportunity in virtualized systems, inside of truly making sure for certain data center workloads that the hot data can be accessed wherever it's randomly stored. Finally, you get to HPC here, big data.
In-memory databases, be it for advanced scientific computing or be it for fast pattern recognition. This is a space, quite honestly, where memory, an insatiable need for more memory with quick access, is driving really probably the most interesting memory subsystem advancements. Here as well, HMC used slightly differently in a way that makes sure that large amounts of data with very low latency can be accessed fast. Getting that memory fabric done right is, frankly, it's a problem that needs a new memory interface, and that's what HMC, some of the technologies we're working on, is enabled to go do. You look across all of this, the question is really, what are we doing about it?
What all of these have in common is, in terms of the silicon participation in these segments, a declining percentage of the silicon penetration going to the actual processor, be it the apps processor or the high-end server processor, and more of the silicon going to the memory, the storage application. By the way, what's happening by virtue of this innovation is that the value of the memory is not just the number of bits. If you think about the last 20 years of the memory industry, increasing densities, generally standards where the interfaces made some advancement, those interface standards were pushed by JEDEC. That functioned as a little bit of a de facto marketing operation. What you have now is that the innovation, the value is in the interface itself, okay?
It's not just the interface, it's the form factor, the particular standard with which we talk to the memory, how we think about the intelligence of the memory, how the memory manages itself. A world of opportunity here. Pulling it back around to what we're doing about that, three technologies here. These are in various stages of development today, development and enabling, frankly. I will tell you that these are big technologies that require a lot of R&D and product investment. The three technologies I'm going to show you today are pretty good proxies. I'm not going to show you all the technologies we're working on. It gives you a good flavor of where we're putting our bets and where memory can do the most good. First of all, in-package memory. Mike spoke about it briefly.
The fact of the matter is, beyond LP4 in mobile, beyond DDR4 with some expansions inside of high-performance computing machines, the data pipe is the problem. There's no good way to increase the speed of that pipe if the first layer of memory is sitting outside of the processor package itself. In-package memory, in one form or another, is getting worked on. Through partnerships, through the kinds of technologies that Micron can bring to bear, with a prerogative of making sure that we get the memory closer to the processing itself. That's something you're going to see more and more of. That's a big architectural trend here that's enabled through things like Through-Silicon Via, which we're in production with.
It's also enabled by advanced packaging technologies to make sure that we can do connections on the order of 1,000 signaling lanes in between the memory and the processor itself. Think of in-package memory, as really the next innovation step along the line. I like to think of it really as what embedded DRAM, was talked about delivering 15 years ago, for instance. We went through the embedded DRAM phase, the fact of the matter is, combining logic and large amounts of DRAM on one piece of silicon was talked about as a big architectural trend for a whole bunch of technical and manufacturing reasons. Outside of a few key applications, it didn't make a lot of sense. What you can do through advanced packaging technologies, again, to get the memory closer to the processor, is a pretty powerful thing, and in-package memory does that.
Second prerogative is getting the processing out closer to the memory. Okay? We announced last time Automata Processor here. This is one way that you can do that, getting the processing much closer to the memory itself. What we're doing here is pattern recognition, in a specialized memory architecture that ensures you get the bandwidth value of memory, the real parallelism of memory to do advanced pattern recognition on binary data, real-time data. It may be networking traffic, looking for virus patterns. It may be bioinformatics, where the fact of the matter is, the kinds of searching, the kinds of graph searching you need to do is best done inside of a large DRAM array, getting the pattern recognition happening at the lowest granular level possible.
A pretty powerful technology and representative, in our minds, of where the big HPC space is headed, ultimately getting the processing dispersed in the lower units and out closer to the memory itself. Finally, storage class memory. Again, here, Scott touched on this briefly. We are looking hard in a, what I would say, relatively advanced development on more than one technology inside of the company that probably the best way to think about it is in between DRAM and NAND, looking for various better ways to slice the memory problem. Okay? Storage class memory is a general term, but what it really says is something that fills that gap. Again, if you think about the historical gap 20 years ago, between DRAM and disk, relatively small 20 years ago, that grows over time. NAND comes in and fills that gap.
Storage class memory is the way, in the future, memory systems will look at filling the gap that's increasing now between DRAM and NAND. There's various ways to think about this. In some ways, you can think of this as the much larger DRAM systems that are made to be persistent so that you don't worry about refreshes. Other ways to think about it is that you take storage systems like NAND, and you move them to this form of memory at slightly higher cost, but in ways that are significantly faster latency, ways that you can get access to the bits in significantly faster time. This is the way that you go and re-architect that subsystem. These are the kinds of advancements that as we think about the four business units that we have, this real continuum of opportunities out there.
Number 1, getting the memory closer to the processor. Number 2, getting the processing closer to the memory. Number 3, re-architecting this memory hierarchy. It gives us an opportunity like nothing, frankly, we've ever seen before. I think of it really in terms of the third phase of the memory industry. The first phase, all about fast advancement and densities, JEDEC-driven interfaces, the Micron of 10, 15, 20 years ago. You think about the Micron of today, really in the second phase, where we've broadened the portfolio through a lot of the early innovation work we've done on RLDRAM, and obviously the mobile portfolio today. This third phase is where all of a sudden memory is the real solution to this level of computing challenges that now exist.
Finally, in terms of the model to pull this off, without a doubt, these kinds of innovations, these are big, expensive endeavors, and frankly, it takes an end-to-end machine to get this done right. Really going all the way from technology to the end application, you need to be thinking about all of the pieces necessary to do this inside of an operation and do it tightly, working left to right here, really on the technology sky, with Scott's team's focus, not just on next-generation DRAM and NAND technology, but now thinking about the packaging technology necessary for better form factors. How do we get these technologies combined together in a way to get the smallest possible form factor size Through-Silicon Vias, for instance? The emerging technologies of storage class memory.
You get to the solutions and engineering side, trying to take those core technologies and turn them into real solutions driven by the customers. Again, we spoke earlier today, not just component design anymore, but the fact of the matter is, a lot of firmware. We got questions last night if the deal we announced changes how we think about our own controller and firmware teams. The simple answer is, it doesn't change it at all, for the simple reason that with the next generation of DRAM, even today's generations of DRAM and NAND-based products, we have an insatiable need for management IP that you do through firmware. Getting the SAS collaboration in place was a great way to say, look, here's one that we're going to co-share.
The fact of the matter is, you think about mobile, you think about PCIe, you think about all of the interfaces coming along behind this and the management of memory, it needs a lot of resources. This is probably, frankly, our biggest area of hiring right now, firmware resources to manage the memory and get the interfaces done right. Third column, the business units. We think this is a pretty good framework. You heard from all of them today. A good way to make sure that we're bridging the gap in between the core technologies all the way out to the customer solutions. Finally, to that point, on the customer side of things, a much more diverse landscape than it's ever been. Darren spoke briefly about the direct-to-business sales force.
That's a pretty valuable channel, not just for the end sales, faster forecasting, but frankly, to understand what the real customers' problems are and making sure that we have earlier access to those kinds of insights. That's what informs the innovation machine. It all really coalesces to make sure that you've got a virtuous cycle here from front to back of getting the next generation of products that truly solves real-world application issues, done in world-class time to market. We like the model. There's a lot of scale benefits here. There's a lot of benefits in terms of the OpEx. We have all of the technologies we need internally to do this. We're not afraid to partner where necessary.
We think it's a pretty good model that has shown the ability to innovate and gives us the best chance forward to be the innovation leaders in the memory field. With that, we'll open it up to Q&A and get Mark and Mark back up here as well. Yeah, Monica?
Thanks for the presentation.
Yeah.
Could you maybe talk about the margin profile of different segments of both DRAM, NAND, and is it the right margin profile for you, and how you're thinking to change the business towards a different margin profile?
Yeah, great question. Probably the best way to think about it is, of course, there's a diverse margin profile, top to bottom, across various sub-segments. Probably more interesting is that there's evolution over time with what that margin profile looks like. Really, the margin profile that we push for is making sure that we are getting our memory pushed to where it can do the most good, where it's valued the most highly. By virtue of these kinds of problems, these kinds of innovations, we have seen that margin profile notably change. DRAM exhibits a good profile of that today, as you heard from both Darren and Mike, working on penetration in the value-added sub-segments in NAND. The higher margin opportunities there are really obvious.
Enterprise computing, for instance, enterprise storage, as well as the next generation of EMCPs in mobile, we believe gives us the best opportunity in non-volatile subsystems to really move the margin lever. Then the next generation of products, as it formed by these kinds of challenges, starts to give the next level. What all of these share in common is this kind of portfolio, this kind of challenge out there, these are all much stickier opportunities, much more custom, much more focused on working with smaller numbers of customers to solve their problems in a direct way that not commoditized at all. We like the model quite a bit. Yeah, Amar.
Hi, thanks. Could you talk a little bit more about the Automata Processor?
Yeah.
It seems very interesting. I'd like to kind of understand. I believe it's more like a co-processor, like a process accelerator, so it doesn't exactly replace processors.
Right.
Could you talk a little bit about what the timeline is?
Sure.
Do you partner with anybody in this area?
Yeah, great question. Automata, this is really representative of what these kinds of architectural overhauls look like. Probably the best way to think about it is as a co-processor. The first instantiation of this will sit on a PCIe card, okay, inside of both networking boxes as well as high-performance computing servers. Through that PCIe card, it's addressed with traffic that's directed towards, by the main system processor, directing the traffic to what this pattern recognition engine does very well. As we've described in the past, what that is really taking a standard computing instruction set and somewhat turning it on its head.
It's a way of saying, look, rather than taking this one thread of computing traffic and going through line by line of code as fast as possible through a next generation HPC processor. Let's take a relatively simple instruction set that can do pattern recognition, it can do elemental instruction, is what we call it, ANDs, ORs, XORs, this kind of thing, and it applies it to an entire array of data in only a way that DRAM can do. It chunks through the data page by page in real time. Every 10 nanoseconds, you will get an answer out of this machine whether you like it or not. That's a level of parallelism that no other processor can do.
To your point, to feed that processor well, you would put this as a co-processor inside of a machine and direct the traffic geared for that kind of a pattern recognition system to it. To enable this, we readily admit, this is not a standard architecture. As a matter of fact, we've put probably just as much effort as we put into the silicon into something called the Software Development Kit, which is really a way of saying the kinds of tools necessary to take a given problem from the networking space, from an HPC space, and make sure that a given application can easily code up their problem in a way that it can be solved by Automata. Frankly, again, this kind of a platform, it relatively nascent today. This has a long enabling time in front of it.
The level of interest from the networking community, from the HPC community, and from areas as diverse as bioinformatics has been unprecedented. Don't think about this as 2015 revenue. I wanted to give you really a picture of some of the things that we've been developing, but this is in sampling form today. Okay, good. With that, we're going to go ahead and turn it over to Mark Durcan for a wrap-up. Thank you very much.
All right. Thank you, Brian, and thank you everyone again for coming today. I've got just a few quick slides to touch on a few topics we haven't covered yet. Then I'll ask Mark to come up and do a little Q&A with me. Let me start by saying, you've heard a lot already today about the things on the right of this page, which are the things that drive our business. You've heard about the importance of products, market segments, markets, developing the right customers, having the right folks to go and talk to them, and the investments we're making in all of those things.
You've heard about technology and the importance of technology to drive those products and drive those future markets, the importance of investing in our business so that we have the right mix of manufacturing capacity out there to be competitive and to deliver those products out into the future. You've heard about solutions and all the things that we're doing for solutions. I'll come back to both the capacity and the partnerships here in just a minute. Those are all the things that drive our business. We don't lose sight of the fact that this is a business, that you guys are all here because you're interested in how successful our business is on your behalf. On the left here, I want to just give you a brief view of how you can look at our business.
One, of course, is in metrics like how are we doing from a revenue growth perspective, et cetera, or gross margin. I think the more important metric for you guys to focus on for a company like Micron, that has a fair amount of complexity in its overall corporate structure is, what is the return on assets? What's the return on invested capital that we're delivering on your behalf? I apologize for all the extra numbers, because the only one I really want you to look at or think about is the non-GAAP number at the bottom there, which is the one that I think is really the most relevant in terms of thinking about our company. We continue to see improvement.
I don't know that that will improve forever because we do need to make investments in our business to grow into the future, we have great growth opportunities. We continue to make improvement. Really, the return, when you think about what this company is doing by leveraging the partnerships and by acquiring assets at the right time, in the right way, and delivering the right products to the right markets and customers, I think is pretty phenomenal, I'll come back to that here in just a minute. I want to talk a little bit about how we're making progress on our capital management. Before I do that, I thought it's worth just quickly reviewing a slide we've shown you before, which is, what are the high-level metrics we're thinking about for the management of this company's capital?
Obviously, we want that ROA to be significantly in excess of our cost of capital, that's currently running about 10%, we continue to make progress on our cost of capital. We'll come back to that in a second. We've said we want to maintain a strong balance sheet. This company has great opportunities out into the future. In the past, having a strong balance sheet has been important to us for a number of reasons, not least of which is it's given us the opportunity to go take advantage of opportunities in the marketplace when those opportunities present themselves. We want to make sure that we don't lose that arrow in our quiver, so to speak, from a strategy perspective. The way we're thinking about our minimum cash balance is this last 12 months SG&A plus R&D plus current debt.
That number currently calculates to about $3.4 billion. I'll show you how we're doing relative to that. We want to continue to make sure we have access to low-cost capital. We want to make sure we keep our leverage ratio in a reasonable range, and we've defined that for you guys as 1.5X. Finally, this is a business that has real growth opportunity that we think is very exciting and that is worth investing in. Having said that, we also believe that over time, this business, while it will fluctuate year in and year out as we think about the greenfield capacity and various technology transitions, et cetera. We think this business will become less capital-intensive. We think we need to exercise discipline relative to how we invest and when we invest.
We have the targets relative to CapEx of sales over the long term. How are we progressing? As I just showed you, last 12 months, non-GAAP ROA 26% is well in excess of our cost of capital. Our capital expenditures over the last 12 months have been about 19%, so we're in that range. When we think about the target capital structure, we continue to make progress there. We did just do a new offering, a number of weeks ago, raised $1 billion worth of cash, which is now on our balance sheet. You can see the cash on the balance sheet is significantly more than that minimum balance. However, I think we've done a reasonable job of articulating for you some of the uses of that on a go-forward basis relative to dilution management as well as other corporate objectives.
Finally, we have seen an upgrade in our debt structure, sorry, credit rating over the last number of months, and we believe that's a positive indication that we're doing the right things for the capital structure. Finally, many of you are also interested in what are we doing to make sure we're enhancing the value of the shares of the company. I think the management team is also very interested in that and works with that aim in mind. Over the last 15 months, we've returned $2.8 billion worth of cash to shareholders through convertible note repurchases. We announced a couple of months ago, a $1 billion stock repurchase had been authorized by our board. While we have limited windows, we want to be able to use that opportunistically.
While that drives limited windows, when we have the opportunity to act opportunistically, we have so far executed roughly $200 million worth of repurchase, and we've repurchased 6.5 million shares plus or minus in that $6 to $7 million range. I don't want to be too specific yet, although we'll certainly report that in the future of our stock. That will continue on into the future as we see the right opportunities in the marketplace. In net, we've reduced 111 million shares, which is approximately 9%, through dilution management for the recent time horizon. Humbly, I will tell you that I think we're doing an okay job overall in aggregate, as we think about how we balance the money we're investing in the business, the money we're using to repurchase shares, and manage dilution in the company. On the partnerships.
We've got a number of very significant partnerships in the business. One of them is Inotera. We've been partners with those folks for a number of years now, and we recently announced that we are or have agreed to restructure what the supply agreement with Inotera looks like going forward. I think it's worth backing up a little bit and thinking about the fact that really in our business, we want to have enduring partnerships. Although the reasons for partnerships can change over time and what they're intended to accomplish can change over time. We want to make sure we have partnerships that are useful for the company and that can endure and change as we move through different periods.
I think it's worth backing up and thinking about the end of 2012, when Micron was in the middle of sponsorship of Elpida and was preparing to bring all that new capacity into the company and sell it out in the marketplace in what at the time was a pretty weak market. When Nanya was struggling in its own business and wanting Micron to take more capacity risk and take responsibility for more of the capacity coming out of Inotera and sell that into the marketplace. We saw great value in that capacity, but we also knew that we needed to have risk mitigation against down markets.
The agreement we put in place at that time with Inotera was that, yes, we will take all the output, but we need a pricing mechanism that protects us, that gives us more risk on the downside in a difficult market because we're also trying to buy Elpida out of bankruptcy. What that market relationship looked like was a market minus agreement that really looked more like an annuity to Micron. It was going to be always positive in terms of the return we generated, but it was going to limit our upside somewhat. That was okay with us, by the way, because in a good market we knew that for Inotera to be useful as a partner for Micron over the long haul, it needs to repair its balance sheet and generate the cash to invest in 20 nanometer going forward.
That's the current relationship, just prior to the new relationship we announced. The new relationship we announced is also a function of what the market conditions look like, but now it's a margin-sharing agreement, which means Micron takes more risk in a poor market, and we make more money in a good market. As a believer in this business and all the things we told you today, we think it's going to be a pretty good business. This is a deal that we think is right for the time, is a more equitable distribution of the rewards of the business that we're helping create with our technology and with our product portfolio. While we take more risk, we think that the ability to make more money, if we execute well from a technology perspective, from a product portfolio perspective, is reasonable.
The folks at Inotera, they're smart folks. They understand the value of the Micron relationship. This really is, in my mind, a win-win. They get to keep Micron as a partner long term with a good set of technologies coming downstream, good set of products coming downstream, and really a stable operating environment out into the future. I think there's a little confusion as to, okay, so how do I model this? What does this really actually mean? The answer is, it depends on how we execute, how good our product portfolio is, how Inotera does in technology deployment, and what the market looks like.
At the end of the day, the better the margin in the business, the more we're going to have and the higher that's going to be relative to the relatively flat slope versus market conditions that we had in the previous agreement. I will tell you that a reasonable place to model this for now in terms of incremental cash margin to Micron in the 2016 year is probably in the 8%-10% range. It could be lower. It could be substantially higher incremental cash margin to Micron under this agreement than under the past agreement. The biggest single lever is, what do you think pricing is going to be in 2016?
To get to that sort of 8%-10% range that I just talked about, really the underlying assumption would be good execution on the technology deployment, because I think that's likely to happen, and maybe a somewhat pessimistic view of what DRAM pricing is going to be in 2016. At the end of the day, the number is much, much bigger if we have flat pricing, and I think that's an outcome that could happen as well. If there are additional questions on that, I can answer some additional a little bit longer, a little bit further on. Singapore. We talked a lot about 3D today. I just want to make one point pretty clear, because we kind of talked around the issue a lot.
When we have 3D NAND technology deployed in our manufacturing fab, it's going to be a significant cost improvement over what we would otherwise be producing, and it puts us on a roadmap that drives significant manufacturing efficiency on a go-forward basis, relative to what could be achievable with planar NAND. It also drives much better performance, like Scott and Darren both talked about. It really is long term where we need to be, and that's why we made the decision to invest in this early, and we feel really very bullish about our technology position here. That's why we're interested in making sure we have the physical plant in place to make the investments to transition our planar capacity in Singapore over time to 3D NAND.
To kind of frame up what that looks like a little bit, if you think about the capacity in Singapore today, it's about 140,000 wafer starts per month. That's in this Fab 10, but it also includes some incremental space in the old TECH Semiconductor fab in Singapore. If we were to transition all the NAND capacity in Singapore, which is our intent, I believe, over time. I think we will leave some capacity behind on planar for a period of time to support legacy applications and lower densities that aren't performance-driven. I don't think that's a big piece of our overall capacity. Our overall NAND capacity is somewhere closer to 250,000 wafer starts per month. This 140 that's sitting there in Singapore, I'm pretty confident over time we're going to transition all that to 3D NAND.
If we go through that process and we convert to our Gen 1, which is 32 tier, then we convert to the second generation at some point while we're ramping this fab space, eventually, the whole Singapore island will be 3D NAND Gen 2. The bit efficiencies that will be driven as we do that are that we'll have roughly the same number of wafers coming out with the addition of this incremental space. We will be able to drive 40%-50% annual bit growth over an extended period of time within this platform, and probably beyond that Gen 2 3D as well, because as Scott pointed out, once you're on that trajectory, now you can add more tiers and scale and generate a pretty good ongoing growth profile. When we do that, we're still going to look at the markets.
We're still going to modulate our investment based on how the technology's progressing, what the customer demands are, what we see as the return on invested capital at any given point in time. In some years, it might be more than this, and some years it might be less. Over time, we can certainly do that, and that compounds to a pretty decent number that'll, I think, be beneficial for Micron shareholders. We undergrew the market a little bit this year, or we will undergrow the market a little bit this year. We overgrew the market a little bit the year before when we took TECH and converted it to NAND flash. We will, as we move through time in the NAND space and in the DRAM space, sometimes oversupply, sometimes undersupply.
In aggregate, the intention is to continue to grow with the market in both areas. On the DRAM side, we've got a picture here of MMT, which is the old Rexchip fab on the left, and a schematic of Inotera on the right. The pink regions are empty fab shell space. I'm showing you the existing footprint in existing manufacturing locations where we could add incremental wafers if we wanted to. The natural tendency, as we've talked about, is as we migrate technology, the process complexity goes up. Going from 25 nanometers to 20, we're losing between 15%-20% of the wafers in some fabs. This space in these two fabs represents roughly a 25% increment to the existing DRAM space worldwide, that gives us the ability to kind of keep up with market growth, should we choose to.
This year, as we've said, we're very focused on technology enablement. We're not increasing wafers, we're growing a little bit slower from a bit perspective in the market. That doesn't mean that's our long-term plan, is to maintain share. Yes, we want to migrate our capacity to more value-added segments, that doesn't necessarily always mean that we're focused on bits. We're not. We're focused on growing our business and doing what's appropriate at the right time based on ROIC. The way we like to set up our fabs, as we talked about in NAND in Singapore and here in Taiwan with DRAM, is we want to be able to add capacity gradually and incrementally based on market conditions. In all these fabs, we've already got manufacturing scale.
We can add to market demand without disrupting the market or without having huge surges in wafer supply that are necessary to reach manufacturing efficient levels. I'm not going to read you all this stuff, hopefully, we've convinced you that we've got a pretty good business here, that we're running it well, we've got a very bright future. We're all very focused on that. We're focused on our markets, our customers, our technology, our people, we're focused on the business and making sure that we make the right decisions as we think about capital allocation for the future. Mark Adams, if you want to come up, I'm going to pitch all the hard questions to you, and I'll take the easy ones. Over here.
Thanks. I have two questions. This is Rohit Shah from Nomura Securities. First, can you guys talk about just your expectation for NAND industry margins, where you think it's going to go? Today, when I look at the numbers, it looks like SanDisk margins or product margins are running in the low- to mid-40% range. Samsung and Hynix are in the 30s, you guys are a bit below that. A lot's changing. You guys are forming new partnerships. There's a big technology transition in front of us. It seems like mix is kind of up in the air. How are you thinking about NAND industry margins and improving, I guess, that spread between you and-
Yeah
the other players in the space? That's my first question.
Okay, there's a lot in that question. Let me start with, I think the NAND market will be volatile over time. I certainly don't want to get in a game of predicting what industry margins are going to be for our business going forward, but particularly for the NAND business, because we have, as you pointed out, significant technology transitions going on. When that happens, it's tough to understand exactly how a technology is going to be received in the marketplace, exactly what the yields are going to be when you ramp new capacity, and exactly what the elasticity of demand, et cetera, is. I think it'll be tough to keep supply and demand matched through time as we deploy these advanced technologies. I think it'll be a good business.
I think margins will be good because I think there's a lot of demand, and I think that there is a disciplined approach to how people are driving this technology, at least as is visible so far. I think that's part one. Part two, how is Micron improving its NAND margins through time? Well, we've talked about the importance of EMCPs and moving more of our NAND into the mobile space. We've talked about the importance of having a better product portfolio in the client SSD space, including TLC NAND as a piece of that portfolio. We've talked about enterprise and some of the benefits we expect to get through the Seagate relationship, as well as other enterprise solutions we're working on internally.
We've talked about 3D NAND and it being what we believe is a significant technology differentiation for Micron, and one that we feel very positively about. We've got a lot of good things going, and we think that over time, that will close the gap. Now, you said something specifically about SanDisk. I think SanDisk is a little bit unique in that they have a significantly higher operating expense than some of the other competitors in the marketplace. A lot of that's driven by the fact that they're in retail, and that market, I think, has been generating very positive margins, but it's starting to slow a little bit.
Where they go, where we go, I think that's tough to predict, but I will very confidently predict that you'll see a narrowing in the operating margin between Micron's NAND business and all of our competitors on a go-forward basis.
Okay, that's helpful. My second question is, I'm still trying to get my arms around the new Inotera agreement. You said, I think, as a baseline, assume 8%-10% incremental margin, but it's price dependent. I was wondering if you could just give us a couple scenarios in terms of pricing and what the impact would be to
Yeah
your incremental margins. Thank you.
Yeah. At flat pricing, you should probably more than double that. At pricing roughly a third of what it is weighted average in the marketplace today, we'd still be making money, but it wouldn't be as good as the current agreement. Does that kind of pound it for you a little bit?
Just the extension of the same question. When you said 8%-10% extra margin, that is Inotera's extra margin.
this extra cash margin to Micron.
Okay.
I didn't listen to the Inotera call or what they had to say. My understanding is they talked about it, that they believe their margins will be 5% lower in 2016. I think they have some margin improvement in their business maybe built into those numbers, or they may have different assumptions as to what they think the market's going to look like in 2016. I wouldn't want to try and narrow that gap. That's a relatively small difference in what could be a relatively large spread of outcomes.
Just a question on the NAND market. If you look at the NAND bit growth, it's been between 35%-40%, 38%-40% over the last couple of years. We have seen kind of oversupply in the market, and all the NAND vendors are still talking about 35%, 40% market growth going forward. You just talked about maybe with extension on Singapore, you could get 40%-50%. Is it possible that we could be in this oversupply mode in the NAND for a long time, and maybe the supply growth needed for the NAND is really not high 30s, it's probably much lower than that, maybe low 30%?
Okay. A few things. Anything's possible. I think that's an unlikely occurrence in the NAND market because I think there is real significant elasticity of demand. I think that while things may get out of balance from time to time, that it is likely to come back into balance relatively quickly, because I think it is likely that companies in the marketplace today will behave as they have been behaving over the last couple of years, which is with a focus on return on invested capital as opposed to market share, where you have five incumbents with really critical mass and significant market share. There was another part to the question, though.
Maybe the supply growth is-
Oh. Yeah. I didn't say we're going to grow capacity 40% to 50%. I said that there's enough clean room space there to support that kind of growth rate over a very extended timeframe. The implication being, I don't think we need to add any new clean room space unless we think that there's a different demand curve going forward after we put this piece in place. We might grow more than that in one year, or we might grow less than that in one year. That's a decision we're going to make as we go. I don't have any reason to believe that as an industry, the incumbents would go out and oversupply the marketplace for an extended period of time.
Yeah, in fact, that was one of the most compelling reasons to do Singapore, because that gave us the ability to modulate. If you go to another greenfield location somewhere else, it would've been a little bit more difficult to get the right scale and the economics to run a profitable business that way. Singapore allows us the ability to add and leverage the administration and the operations in place and scale appropriately to the market demand.
Two question here. Going back to the Seagate arrangement, it seems to me there's more into it. It's definitely more than just getting some forecast on SAS demand. Can you elaborate how this whole arrangement or alliance came about? Seagate has historically had a relationship with Samsung, you're coming in. What are the risk or reward here? Again, I'll go back to, it's got to be more than just getting a forecast on NAND demand for SAS application. I have a follow-up.
Let me answer, maybe Mark wants to add something. I don't think I want to speak for Samsung or Seagate about that relationship. All I can say is apparently it wasn't working out to either party's satisfaction, because I don't think that they've been doing a whole lot together recently. I think we've done a few relationships. I think we understand that for relationships to be successful, both parties have to get something they want and need, and that it has to be a cooperative and collaborative relationship over time. There is a lot of stuff in that cooperation and collaboration that's going on here.
As Darren talked about, there's a lot of collaboration going on with knowledge about how the NAND works and what knobs to turn in the firmware, and how to think about controller functionality on a go-forward basis to make sure that the whole thing can be optimized on a go-forward basis. There is right to supply that Seagate needs in order to grow its business. There is a second source in the marketplace that customers want to feel comfortable that they can buy these drives in large volumes. There is access into the enterprise market for Micron with a drive and a partner that has a history and reputation for delivering quality drives and processes to support them.
There's a lot of things around there that make a pretty compelling package that create a lot of extra value and maybe a significant share in a profitable and rapidly growing market segment that I think creates a pie that is big enough that it's pretty exciting for Seagate and it's pretty exciting for Micron.
Sure. Follow-up has more to do with the near-term business. During the November conference call, you talked about stabilization in NAND prices. The DRAM guide came in a little bit worse than expected. How do you see market that has evolved since, and how do you see the inventory that could be a factor after Chinese New Year holiday?
I'm going to let Mark talk about market conditions right now. I'm not sure I understood the first part of your question was relative to something somebody said when?
Going back to the November earnings conference call
Oh
where I think Mark suggested that NAND prices have actually started to bottom out.
Yeah.
The DRAM prices were a little bit slightly weaker. It seems like even DRAM has bottomed. Maybe if you could provide some update
Yeah
on how you see inventories at a current environment being a factor post-Chinese New Year holidays.
Yeah. Mark, you want to?
Sure. I think how we look at it is that the inventory situation could be a combination of factors, one of which is when competitors or companies are actually making transitions and they have lower yield product or lower spec product, what have you, that hits the market price, that can have a short-term impact on market conditions. We're not of the opinion that inventory is way out of whack post-holiday, and quite honestly, we think that seasonality-wise, this is somewhat to be expected. You could argue about timing and what have you, and we're not in the business of commenting on that, but we're not overly concerned about the directional signals necessarily as a trend to go on forward. We're not commenting on that.
We do think that this is more seasonality in play and potentially some intermittent supply that hit the market for technology ramps and processes, and we feel pretty good. It's interesting. You're talking about one part of the market, and we're seeing really strong demand signals in other part of the market, which was my earlier comment today. Some of our segments are still very much in constrained mode, and some of our segments are experiencing the dynamic you're talking about today. Overall, still pretty good.
Blend basis for both NAND and DRAM are kind of stable?
Yeah, we're not here to update our guidance that we did give.
Thank you.
Melissa? No. John. No, go ahead.
Yes, this is Srini from Summit Research .
Yep.
My first question is, what are the gross margin improvements that you might say the investment community is not appreciating at this point? What are the initiatives that you're doing on gross margins?
In NAND or DRAM or both or?
Both.
Okay. I think I already addressed some of the things that are going on in NAND and the positive trends across a number of different segments, and probably enough said on that other than, I think you'll notice it as we move forward quarter by quarter, and I think it'll be significant by the end of 2016. You'll take a look at it and you go, "Hey, that's pretty good." DRAM, obviously, we just talked about the Inotera relationship and how that might play out or not play out. Additionally, there's nothing like a well-executed technology transition to help drive margin in a manufacturing business. Having said that, be cognizant of the fact that as we spend capital, it takes a while for the capital to be installed, qualified, processes qualified, wafers loaded, run through the fab, ramped up, et cetera.
These things don't happen instantaneously, but I think over the long haul, as I've said for a long time now, closing that technology gap on the DRAM side so that our weighted average deployed manufacturing capacity is more similar to that of the other two DRAM manufacturers, I think will be positive for us driving gross margin.
I have a quick follow-up. How can you give me confidence that your NAND plan is going to come to fruition?
You take some?
Yeah. I talked to a few tables last night at dinner. If you go back and track what we articulated as our recovery plan two quarters ago, the elements are the same. In fact, what you heard today was our mobile NAND business shaping up pretty well. What you heard today is that TLC enablement are meeting the milestones we articulated in the past. We're quite excited about not just the announcement today that was in Darren's business, but some of the progress we're making on PCIe and enterprise storage, and in Scott's section today. A lot of the element themes that we talked about half a year ago about what the recovery would take are actually playing out as we predicted, as we called. We're extremely confident about the recovery.
Yeah.
Mark, a couple of questions. First, your comments on Inotera were helpful, for those of us that are still modeling-challenged, how would fiscal year 2014 have looked if the Inotera agreement were in place for the fiscal year 2014? How much more accretion to the model would there have been?
I haven't run those numbers, John, it would've been significantly more than the number I mentioned.
That's helpful. As a follow-up on the
The other thing is to keep in mind here is supply, right? What is the total bit output of Inotera in 2016 on 20 nanometer versus what was it in 2014 on 30 nanometers? You've got more bits you're earning margin on. What's going to happen with ASPs? It's a very dynamic model.
Got it. My second question is just on the DRAM market in general. Last year, Samsung grew bits a lot faster than everyone expected coming into the year, and they gained market share. I am kind of curious from your perspective, is there a market share threshold that you don't want to fall below in DRAM? Then maybe as you address that question, there was an interesting one-liner on one of Scott's slides that said that the 1X transition was a greater than 30% cost reduction versus 20 nanometer, which seemed a lot more than I would have expected. Is there something going on with 1X that it is going to be a very efficient node for-
Yeah
Micron or the industry?
Yeah. I think it is actually more about Micron internal dynamics than anything else, in terms of how the cost reduction going from 30 to 20 to 1X, how that all plays out. Sorry, the first part again.
Is there a threshold to the market share?
Yeah. Yeah. I wanted to make the point today that we're not really interested in giving up share over the long haul. We've got a good business. We've got customers that want to do business with Micron. While my primary filter continues to be return on invested capital, I don't want this business to go into decline as I try and drive higher and higher ROICs. That's a little bit of a balancing act. One thing you should notice about Micron is we've done a pretty good job maintaining revenue share, and that's really how we think about the business. It's more about revenue than bits. Having said that, I think this company needs to have the critical mass to be significant to the significant customers out there in the marketplace, and that requires having some significant bit market share as well.
We'll look at all those factors through time and make the right decisions.
I guess a little follow-up on that, because that was pretty much going to be part of my question, was your bit share versus revenue share. Can you maybe give us a little bit of what's going on underneath the covers of that? You have maybe ceded several points of bit share in DRAM in the last couple of years, but yet I don't hear you talking about what's really gone on that's enabled you to perform so well on the revenue share side.
Okay. Let's talk about bits first. In 2012 and 2013, we took TECH offline, converted it from DRAM to NAND, and we did that for a number of pretty good reasons, I think. That meant that we grew less than the market in that timeframe. This year, we're very focused in 2014. This year, we're very focused on technology transitions, right? We've got a certain amount of capital, and we're allocating it to a lot of different things that I talked about. The stuff that we're investing in the business is about closing this gap to drive margin to the right place and to make sure that we're seeing the right economics relative to our competitors. That's why we're focused on that today.
In the future, I think we can take a different approach, and I wanted you to understand that I have the ability to add bits incrementally in the existing floor space. How have we done such a good job on a revenue basis as we ceded some bit share? Because of the decisions we made. We spent a lot of the day talking about it. We spent a lot of the day talking about the progress going on in the mobile business. We spent a lot of the day talking about where we've been bit constrained in servers, and that's a growing market. We can actually go into that more now that we have more capacity than we did pre-Elpida merger. We've done some great work in Jeff Bader's embedded business. The networking business is phenomenal and has more upside.
In all those market segments where Micron is focusing and working closely with customers to deliver differentiated value, we're having success. That's, by the way, another reason why we feel comfortable that we'll solve the NAND gross margin gap here over the next number of quarters.
Great. For my follow-up, just a really quick one. CFO search, can you give us progress or a timeframe at which you might think that it'll be completed? Thanks.
Yeah. I don't want to set a timeframe. I've talked to a lot of good candidates, and I've got candidates that I'm comfortable with. I'm still talking to more. It's an important position, and I think the company deserves or expects that I go out and do a lot of diligence and bring in the absolute best candidate I can. I don't feel in a rush. We've got a very solid finance team, from treasury to controller, to region control, to tax and GAAP report. We're in great shape on the finance team. I want to get a real superstar for CFO, and I'll do that on the timeline that presents itself.
I think we're out of time.
Okay. All right. Thank you all. Hopefully, we covered most of the things you wanted to talk about.