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, and thank you everyone for attending Micron's 2016 Financial Analyst Conference. I'm going to give a quick kind of overview of the day before we start. We're going to start off with Mark Durcan up first to talk about the industry environment and some of the key strategic priorities for the company over the next year to two years in particular. Second up will be Scott DeBoer, updating our technology execution and roadmap, including current and next-generation technology. Number three, Brian Shirley will come up and talk about our memory solutions overview, essentially tie in all the pieces of technology, operations, product portfolio, and some of the key end markets. Fourth, Ernie Maddock, our CFO, will come up and talk about our financial and operational performance. Lastly, Mark Durcan will come back up to wrap up the day.
Just so everyone's aware, we'll do Q&A at the very end after Mark is finished. Please hold your questions until then. You might want to take note as we go through the presentation, you can make sure to come back to that. We will have all the presenters back up to take questions at the end. With that, I'll turn it over to Mark.
All right. Thank you, Ivan, and thank you all for coming. It's always good to see so many familiar faces out here in the crowd. I think we've got a good presentation, as Ivan just referenced, for you today. A lot of good content. What I will do is just try and set the stage for the detail that's going to come from the rest of the team later on. Without further ado, many of you have seen this slide in the past, but I think it's worth just sort of refreshing. Micron plays in a broad swath of memory technologies, not just components, across both volatile and non-volatile memory. Roughly an $80 billion addressable TAM for our products, in what arguably is one of the better long-term growth pieces of the semiconductor industry.
Semiconductor industry, of course, leveraged to global GDP, but also some significant growth drivers embedded in the memory piece that I think give us the opportunity to outperform the rest of the semiconductor industry, even in a time of potentially muted growth. I'll come back to that here in just a minute. Obviously, here when we talk about the memory market today, we're showing you just memory technologies that are of significance today. We do believe there are emerging technologies and technologies that we've recently announced that will become a significant piece of the memory market, and myself, Scott DeBoer, and Brian Shirley will all address some of those opportunities as well that we believe are, at least over the next four or five years, significantly additive to the overall memory TAM. Looking at just the big pieces of Micron's business today, NAND and DRAM.
When we look at the opportunity for non-volatile memory, primarily on a go-forward basis, 3D NAND, we see significant growth drivers in terms of NAND bit demand. It is a very elastic market, and I think as many of us look at the market today, we can see that we're starting to hit the tipping point in different segments of the market. Today, corporate clients, I think we're already sort of in that tipping point where things are starting to snowball and more and more of corporate clients now are shipping out with solid-state drives and solid-state memory. That's coming very quickly now as costs come down with 3D NAND transition into the consumer space. The enterprise, I think, is bound to happen.
It is happening, and it will continue to accelerate, probably in ways that are not even clear to most of the customers today as new enterprise solutions roll out and as we continue to develop the types of solutions that really take advantage of the NAND media as opposed to the hard drive media. Early introduction of NAND into storage has been really positioned as replacement for hard drive media. We're not getting the full value of that technology in the enterprise yet, but I think as we move through the next number of years, you're going to see that, and you'll see the acceleration and adoption that comes with that as well. All told, for the NAND market, we're showing you here approximately 40% CAGR in terms of bit demand on a go-forward basis. I think that's conservative given the elasticity in the marketplace.
I think it could be significantly more than that. The question you always have to ask, well, demand at what price or what cost point? We think that there is significant opportunity to upside this demand, and we're going to prepare ourselves to make sure we have the right solutions to address that demand. On the DRAM side, we think that the demand growth is actually in approximately the mid-25% range. I'll come back later with sort of a market segment by market segment view of that, and you'll also have some more from this on Brian Shirley. We think that there are significant demand drivers for DRAM memory on a go-forward basis, notwithstanding some of the other new memories that are going to enter the marketplace.
We believe that notwithstanding slowing unit growth in some of the end markets, there's enough demand for system main memory, scratchpad memory to continue to drive significant DRAM demand growth well into the future. Obviously, on top of this, I'll come back here in a minute and I'll show you a little slice of one market segment that can be addressed with 3D XPoint memory. That 3D XPoint memory is one of a number of different new memory technologies you'll see in the marketplace over the next few years. We think have big potential that at least in the short term, again, is significantly additive to the picture I'm showing you here relative to DRAM and NAND. Looking at supply, you can see the long-term trend here to lower supply growth as technology migrations have slowed and as end markets have diversified.
We think on a go-forward basis, DRAM technology supply is in the low to mid-20% growth. In fact, I think as we think about 2016, it's hard to know exactly if it's 20% or 25%, but it's in that range. As we look to 2017, we think it could easily be less than 20%, depending on what happens relative to new supply additions. This trend is very real. It continues. Hard to know exactly what would happen with supply, but as we monitor what we know today, we're not aware of significant new supply additions coming in the memory space. We think that what is a slightly oversupplied market today will continue to improve as we move through the year and potentially becomes more like what we would expect for the long term as we move into 2017. On the NAND side, again, significant demand drivers.
We'll come back to the detail on that. Supply growth really here are going to be driven by what incumbents in the marketplace invest, because there is no way that technology migrations alone will keep up with the demand growth that we're going to see in non-volatile memory. Our view is that we have a relatively slow year this year as people transition to 3D NAND. We could see that grow a little bit as we move into 2017. The long-term rate of supply addition is not going to keep up with demand. We've seen significant penetration of TLC already. That trend has kind of played out, and while we may get more bits per cell eventually than just three, that trend is a longer-term trend.
In the near term, there are significant wafer capacity that has to be put in place in order to continue to drive supply to meet demand. I think maybe a little bit more of a volatile situation, but certainly a beneficial situation relative to non-volatile memory on a go-forward basis. 3D XPoint. Brian Shirley's going to come and talk to you a little bit later about all the various applications where you might think about memories like that going in the future. If you just pick one small slice, which is mainstream servers, where people might envision using a 3D XPoint memory for in-memory databases to have access to significantly more data close to the processor, and in a format that is a non-volatile, but with very fast read latency.
We tend to take a conservative view on these kinds of things because we will need to deliver to what we show you. This could be a small forecast relative to a small piece of the overall addressable market. We do think that this type of memory will be significant in the server market and will drive really over the near term, four or five years, significant growth in the overall memory market as it becomes much more compelling to add this kind of memory into the server space at the cost points that can be achieved and the power thresholds that can be achieved with a memory like this. Okay. Our end markets continue to diversify.
If you look at the progression of DRAM from a largely PC market in 2000 to one in 2019, where less than 20% of the bits will be going into the PC segment. Really a very virtuous trend that continues out into the future in terms of the diversification of where our products go. As you look over to the right and you look at, okay, what's the CAGR in those various markets for DRAM bits? You can see that we think that there's good growth far in excess of what can be provided through technology migration in all of those large markets. If you think about a world where we can maybe grow bits 20%-25% on an annual basis through technology migrations, that trend, by the way, is long-term towards even slower growth.
We continue to see demand in server and storage that's in the 40% range, in automobiles in the 40% range, even in smartphones in the 25% range. Very strong demand in those end segments for DRAM. The only one really that we think probably underperforms relative to what we can do through technology migrations is in the client PC. Looking at NAND, similar situation, sort of the low-end markets, cards, consumer, removable storage. That market continues to shrink, and is in long-term secular decline relative to the overall size of the market. The markets that are growing, the solid state storage markets and mobile markets, continued robust demand growth for NAND memory in those markets, as well as in server and storage. Overall, we like the fact that our products go in all these end markets.
We like the fact that our products can be differentiated in those end markets, and that in those end markets, customers more and more want a differentiated solution with control functionality, firmware and software wrapped around it. This really is the future of the memory business. It's not a commodity component business anymore. It's memory solutions to end segments, Brian Shirley will talk more about that in a minute. What are we focused on as a company? It's a lot like what I told you last time we had an Analyst Day, that should be the fact. We shouldn't be chopping and changing our priorities all the time. For 2016, it's really about execution of technology development and migration.
Thinking about in the DRAM piece of our business, it's really all about continuing the effective ramp of what we call 100 series DRAM or 20 nanometer DRAM in our fabs in Japan and Taiwan. Scott DeBoer is going to give you an update on that, and it's about the 1X node, which is now running in our pilot line in Hiroshima, will be deployed later in the spring into Taiwan, and making sure that that also goes well. On the non-volatile front, it's all around the 3D NAND ramp. We've got updates on that technology, what it looks like, why we're so excited about our relative competitive position. By the way, I'll tell you, I'm very excited about what's going on with our storage roadmap as well, Brian Shirley will have some comments on that in a minute.
Couldn't be more happy about where we are on NAND today relative to the technology and the way that's coming together with our solutions roadmap as we move through 2016 and into 2017. The priority for the company is keep our eye on the ball, make sure that continues to go as smoothly as it has up until now, and then layer on top of it the 3D XPoint market enablement to make sure that we have the growth in multiple market segments that we see as possible and exciting for the future. Finally, again, this business on a go-forward basis is about memory systems. It's not about commoditized components.
We continue to focus on improving our capabilities relative to all the things that go into delivering system solutions to the marketplace, whether it's controllers, firmware, software, other hardware enhancements in terms of packaging capability, appliance design, and the coalescence of all the different memory technologies that really only Micron can bring to the market today. When you think about who can really deliver this, it's Micron, because we have advanced DRAM, we have advanced NAND, we have NOR, we have 3D XPoint, and we have a lot of other interesting technologies in the pipeline. Very well-positioned for the future. Ernie will talk about what this means from a financial perspective to us and how we support all these priorities on a go-forward basis. With that, I'm going to turn it over to talk to Scott.
I think he's got a great presentation for you on how these technologies are coming together and how they're rolling out in manufacturing. I'll be back at the end to take your questions. Thanks.
Thanks, Mark. I'm going to start off today and talk a little bit about the technology deployments last year, so largely focused on about a year ago when we were in the same place, talking about what our priorities were. Clearly around 20 nanometer DRAM and our first generation of 3D NAND technology. Later in the year, we talked to you about 3D XPoint. Through this year, 20 nanometer DRAM went into both Japan and into Taiwan, and I think we're very pleased now with where the technology execution is, where the manufacturing execution is. Really moved forward in a positive way, and I'll talk about where that status is headed. On the 3D NAND technology, that's deployed into Singapore, first generation, and then into Lehi, Utah, with the 3D XPoint.
I think you'll see a really positive story here around the technology deployments that we were focused on last year, and some good news on several fronts. First on the DRAM technology progress, we've focused a significant amount of our resource over this last year in executing on the 20 nanometer yield ramp. On this particular slide, you can see the yield execution through the year, and I've put on here the comparison to our 25 nanometer node through the same period of time from technology introduction through what's today on the 20 nanometer. You can see a couple of things from this. One is that effectively after last analyst meeting, kind of end of summer, early fall, we did have a period of time where our combined infrastructure yield ramp was not increasing for a couple of months.
The other thing you'll see, though, is that things now, as we're rolling out the real manufacturing output, are ahead of plan. The combined 20 nanometer yield is significantly better than even we accomplished on 25 nanometer. Also on this particular slide, you can see where we're headed on bit crossover. The bottom graph is a comparison of our leading technology nodes. The 25 nanometer DRAM relative to the 20 nanometer DRAM. You can see we're crossing over with what is effectively now in terms of the bits produced on 20 relative to 25. We're on track to have more than 50% of our fab bit output on 20 nanometer by our fiscal Q3 of this year.
In addition to the cost benefit of this node that we've talked about and the benefits coming up throughout the rest of this year, this node is critical to us for the higher density enablement in a variety of products that are all coming out on the 20 nanometer node to enable our mobile business, key eight gig parts on compute and across network and some other areas that Brian's going to talk about. We have a significant product portfolio enabled now on 20 nanometer, and you'll see some benefits from that through the rest of this year. On our 3D NAND first generation, a couple of key things.
One, last year at this time, I talked to you about how we had some unique architecture benefits to our 32-tier technology relative to what you'll see through the rest of the industry, and that all 32-tier technologies aren't created equal. Today to talk a little bit more detail about that, one of the aspects of our technology, on the right of this particular slide, you can see the structure of how the memory is built, how NAND is built. Traditionally, in planar NAND and also in 3D NAND products that you've seen to date, the CMOS circuitry that supports the memory array is outside of the array. The overall die is made up of CMOS circuitry, which we call periphery, in combination with the array.
The unique thing about the technology that we have rolled out is that we have taken the CMOS circuitry and substantially put it directly under the array so that it looks like you see on this picture. Now, what this does is effectively gives you significantly better bits per wafer performance and a better cost structure. When we talk about our 32-tier, it's actually substantially closer to what other people would talk about in terms of a 48-tier product. This ultimately gives us about a 20% cost benefit relative to if we put it outside the array, so in terms of shrinking the die size. In addition to some unique architecture developments or capabilities of our NAND technology, we also have continued to improve the raw performance of the NAND technology, and we're really pleased with where that has gone since we started the rollout of the technology.
From the beginning, we've talked about how this 3D NAND technology is going to really be significantly better than planar technology in terms of performance, in terms of endurance, the number of times you can read and write to it, and that it was going to be a real step function in performance relative to planar NAND. What's happened through time, and indicated on the bottom graph here, is even beyond our original expectations for the performance of 3D NAND relative to planar, we've made another step function in terms of the performance. Now see the capability as being substantially better, which Brian will talk to, some of the products this helps enable. It really pushes the performance to enable enterprise-type products and other high-performance type products. Overall, really some great work here to get 3D NAND to the finish line.
In terms of the progress on the manufacturing rollout of the first generation of technology, from the beginning, we talked about we've prioritized our TLC roadmap over our MLC roadmap. On this graph, a couple of things to highlight. One is that because of this prioritization, the bottom right graph shows that our yield is basically a parity between TLC and MLC products. The top shows a picture of a wafer map of yield on 3D NAND, and we continue to outperform our yield plans in manufacturing relative to what the expectation was. Remember, this is very disruptive technology compared to anything that's run in a memory manufacturing operation before, and a lot of the detail that went into this upfront is now paying off in terms of some really phenomenal yields, and we're really pleased with that.
Another unique thing I think about the technology, partially back to what I talked about before, which is this ability to put the circuitry under the array, is that we do get a significant cost reduction from our first generation of 3D NAND relative to planar technology, more than 25% cost reduction for that first generation. In terms of the technology we've rolled out, we're not rolling out a technology that's at parity with planar. We're coming in directly with an advantaged product line. The bottom graph on this shows that prioritization of the TLC ramp, and I think you can see that based on that prioritization, the TLC bits out of our manufacturing operation in 2017 will be dominated by TLC, with the crossover coming towards the end of the fiscal year. Switching to the status on 3D XPoint.
We talked about, in July, the introduction of this technology. We're making good progress on yield and manufacturability. As you can imagine, a technology as different as this, there's different kinds of things that have to be worked with. We're very happy with the progress, and we've rolled out sampling with a significant number of our select product partners. That's going well, really with a focus on enabling end solutions. Overall, the schedule for this technology, the volume ramp is on track. It's really, for Micron, a story of revenue building, as Mark had showed, in 2017 and 2018, and growing over time. 2016 is more of an enablement year for us. Okay. Shifting focus to where our priorities are for 2016 and beyond. On the DRAM side, we continue to be very focused and aware of our technology position relative to our competition.
As I described to you last year, we have, post the Elpida acquisition, put substantially more resources on enabling our DRAM technology and getting it to a much more solid position. This upcoming year, the 1X nanometer deployment will be the focus, as Mark alluded to, I'll give you an update on that one in some more detail in the next slide. In addition to that, we have substantial efforts on the two nodes beyond that, and are much more confident about our ability to do parallel development and enable nodes quicker. On the 3D NAND side, the focus is all around the second generation enablement, which is going well in R&D right now, it'll be a 2017 story from a revenue point of view, an output point of view. Behind that, we have multiple generations of continued scaling of density and cost reduction.
On the new memory side, focus, as we talked about, really on 3D XPoint. We have some other interesting things I'll allude to a little bit on the other types of new memory that we're really excited about. On the 1X nanometer deployment, our focus through this past year and still is on the yield ramp in Hiroshima. If you remember, the 1X nanometer node is really the first node for the combined Elpida and Micron team together to put together a technology from the ground up. There's a significant amount of enthusiasm and focus amongst the team on this first combined node coming out, and really providing a significantly better benefit than what we've had in the past on a node-to-node transition.
If you look at the density projections on the bottom right graph, we're getting a significant benefit, and we've talked about the cost reduction on 20 nanometer. The 1X nanometer is targeted to provide an even better cost benefit, and we're targeting significantly better than 20% cost reduction in moving from 20 nanometer to 1X nanometer. This focus shifts to Taiwan, as Mark mentioned a couple of minutes ago, actually in the next month and a half, in March-April timeframe. We'll be focused on the deployment and execution of 1X in manufacturing through the rest of this year, fundamentally in Taiwan. On the next generation 3D NAND side, again, I think a really positive story here. This technology is yielding in the R&D fab now, with a transition going on to the Singapore manufacturing fab.
Definitely on track with the aggressive development schedule we set at the beginning of the program. As we talked about a minute ago, or I talked about a minute ago, we're getting a decent cost reduction on the first generation relative to planar. We get an even bigger cost reduction, again, in this case, on the second generation 3D NAND. It will be a very good story for us in terms of density and cost reduction enablement. More than 30% relative to gen 1. This summer, we have our Singapore expansion for NAND coming out with first wafers on gen 1. Shortly after that, also in the summer, we'll have the first wafers in this tech deployment coming out of the new fab in Singapore also.
If you look at the bottom graph on this, you can get a feel for how our bit crossover relative to the prior leading technology progresses over the next couple of years, with the comparison of our first generation 3D NAND relative to 16 nanometer planar happening in fiscal Q3 towards the end of this year. Beyond that, a couple of months later, it'll cross over versus all of our planar NAND. It crosses over with 16 nanometer planar leading-edge technology node first. Not shown in this graph is just a crossover a couple of months later with all of our NAND bits. In fiscal 2017, towards the end of the year, you'll see significant volume coming up on the second generation of 3D NAND technology, this technology will be crossing over towards the end of fiscal 2017. Okay.
On the future memory technology side, we talked about the current status of 3D XPoint. We're also very excited about the roadmap that has been under active development over several years, but that's really coming into focus now for both significant cost reduction over multiple nodes of 3D XPoint, as well as some nice performance benefits in terms of power and speed that come with the future generations of 3D XPoint technology. Behind that, we have focused programs on other new memory technology, really aimed at the highest performance end of the spectrum and the lowest cost end of the spectrum. We have some really exciting technology advancements in the R&D fab right now on both of those two extremes, and are working with some key customers on enablement of other technologies. This continues to diversify the overall memory technology landscape.
In conclusion, very focused on DRAM competitive position improvement. I think with the 20 nanometer rollout this year, through the rest of this year, and the 1X nanometer technology coming out, I think you'll see substantial improvements in our DRAM position, leading to significantly better business opportunities for us there. On 3D NAND, as Mark said, Brian's going to talk about in quite a bit of detail, we have a great product portfolio coming on 3D NAND. The technology is solid. I think this is a real bright spot that the entire Micron team is really excited to put out there over the next year. 3D XPoint, talked about enablement, really a story of revenue in 2017 and 2018. Didn't talk about it today, probably next year, we have a significant amount of activity on the advanced package development side.
We had rolled out Hybrid Memory Cube and other technologies that we've talked about in the past. We have some exciting new developments to talk about in the future. To support all this activity, we've done a substantial expansion to our R&D infrastructure for technology development that comes online towards the end of this year. With that, I will turn it over to Brian.
All right. Thanks, Scott. Good morning to everyone. For those of you that I haven't met, my name's Brian Shirley. I oversee what we call the memory solutions developments at Micron, taking the processes, the manufacturing technologies that Scott's group's working on, and putting them in the form of higher value products, specifically for our key customers and partners. What we want to show you today is really a picture of two things. First of all, the opportunity out there, secondly, the ways Micron is taking advantage of that with our end products. The big story here on the opportunity side, no secret, but it is just an unprecedented explosion of data. That is driven, obviously, by the creation side, specifically with mobile and client, what's happening with video, with applications, with imaging.
It's also driven by the Internet of Things, a nascent true automotive computing market here that's generating gobs of data. What happens with that data next is getting it moved through the networking infrastructure and getting it up into the cloud, the big data, the enterprise space. Across every single one of these applications, what's happening is a need for true system evolution. The systems out there today are just simply not capable of dealing with the sheer amount of data being created that needs to be moved, that needs to be stored, and needs to be acted on. What that means is a call specifically for very much innovation on the memory subsystem side. What we're seeing, what our key customers are coming to us for, is a need for breaking through these particular challenges.
These challenges, this opportunity out there, the world is no longer compute bound. The world is memory bound. For a memory company, that puts us in a pretty key position. Across the board, calls specifically for innovation in our products that deal with how we treat this data, specifically the hot data that needs to be acted on quickly, the power consumption that comes along with that, getting what previously was treated as cold data closer to the processing system, the various form factors, meaning the sizes, the location close to the processor, form factor within a mobile phone, for instance.
Increasingly, challenges around hiding the physics of the memory, things that we've already started dealing with in non-volatile memory, the number of times that you can read and write a particular bit, hiding the physics from the end system or making those manageable, guaranteeing the reliability for the application and getting this done in a way that is affordable and cost competitive for the end application. These are steep challenges. Like I said, these are challenges within our wheelhouse. What it calls upon for, though, is a memory company that has a number of things at its fingertips. The first, obviously most critical, being leading memory technology, a portfolio of technologies, not just the world of 10 years ago with just DRAM shrinks, but increasingly the non-volatile portfolio, the emerging memory technology portfolio, and ways to put those together in various combinations.
That's really what I mean by advanced subsystem execution. Putting together these technologies in ways and getting them managed through firmware, through controllers, through software. Even the packaging technologies, as Scott alluded to, eMCPs and some of the form factors that are happening on the compute side, that put these in a form usable by our customers. Finally, the ecosystem access, meaning the access to the engineers with our partners, where the real system innovation is happening, the system labs out there, making sure that on a go-to-market strategy, we're able to get these products designed in a way so that they are truly valuable and usable by our customers. What I'm going to do next is walk you through really four of the most critical macro segments, roughly corresponding to how we've organized our group on business unit side.
We'll show you some of the key trends happening in each of these spaces, speaking specifically to this kind of innovation and what we're doing about it. Compute and networking side, the big story here, traditionally, 10 years ago, the old client market has really transformed here. What this is today is a story about the enterprise and cloud growth. What we've done here is shown you a multi-year CAGR going forward. You can see from these two CAGRs, this is a market that continues to grow. One, it's a market that is close to actually about a third of the overall DRAM market today and poised for continued growth. By the way, this kind of memory, generally going forward, it needs to be DRAM. It values the latencies, it values the kind of close access to the processor, the fast write times that DRAM enables.
The story of enterprise is scale-up architectures, more components within the system to enable high-performance computing, and the story within cloud is scale-out architectures. Even there, high-performance computing coming in, but happening more across a dispersed system with multiple nodes. In networking, again here, relatively smaller segment in terms of absolute memory bits, but a segment that's really been transformed in terms of what it needs from memory to actually go and get the next-generation systems designed. Continued push on the endpoints out there, as well as the kind of systems that can actually truly move the data speeds coming at it. Our portfolio, specifically on the compute side to deal with this, enterprise cloud, this has been a good segment to us over the years. 20-nanometer DRAM here, specifically the 8-gigabit products getting out to market, DDR3 and DDR4 these days, very cost-competitive technology.
A bigger story here, though, the introduction by Micron of something called non-volatile DIMM. This is an industry-standard form factor memory module that actually has both DRAM and NAND on it. It's really the start of persistence in the compute world. A memory module that functions as scratchpad memory, but with NAND on board to take care of system checkpointing, of journaling, of system reliability, receiving just a ton of good interest here from our key partners across both the SoC space as well as the software vendors. Moving over to networking, a space that's been good to us over the years. Good leadership here, specifically with RLDRAM continuing, reduced latency DRAM. This is a customized architecture specifically for networking. No other way to get good next-generation systems designed with the kinds of latency needs they have. HMC for both enterprise as well as networking here.
As you saw, one of our key partners this year with introduction of this next generation of a co-processing high-performance SoC device based on a variant of HMC, something that we call MCDRAM. Micron is shipping that today. In the networking space, specifically, the true next generation of networking systems based on HMC, shipping in volume today. On the graphics side, here we have just a push for the ultimate in high speed. The world has transformed to GDDR5 today. If you remember, a couple of years ago, with our acquisition of Elpida, we integrated a graphics portfolio back in to our DRAM portfolio here, and this has gone extraordinarily well. As a matter of fact, we're pleased to announce a world's first, something called GDDR5X.
This pushes the next generation of graphics speed upwards of 12 to 14 gigabits per second and really gives next-generation graphics cards and consoles the pure bandwidth that they need for today's systems. Embedded in storage, three sub-segments here, all of which have pretty good CAGRs and speaks to a lot happening on the memory side. Automotive, really, as a lot of you have been hearing, the big push here, frankly, across entertainment systems, autonomous systems, ADAS systems, is a need for turning the automotive platform into a true compute platform. What that means for us is pushing on leading-edge components. Traditionally, five, 10 years ago, if there was any memory used inside of an automobile, it was generally old generation. These days, things are pushing on leading-edge densities and technologies. Industrial multi-market, the story here is just very, very strong unit growth.
These are not necessarily, on average, large compute systems, but across the entire distribution space, a lot happening here to take the sheer number of units up. Finally, on the connected and home consumer space, the story is 4K. Video's pushing things hard. As many of you know, the 4K adoption rate has been spectacular. That needs a lot of memory inside of DTVs, inside of set-top boxes. Our positioning here is very, very strong. As a matter of fact, in automotive, we have a top market share, and we have put in place the steps to keep it that way.
Very, very good partnerships, a good field lab structure, and most critically, the products that the automotive partners are looking for, specifically long-lived DRAM and NAND portfolios, and really pushing the leading-edge designs out there, 20-nanometer designs, that will be in production as long as necessary to make sure the automotive partners get the computing systems designed in with the longevity they need to be successful. On the industrial side, a good distribution structure here that's really helped us out, getting our components out to market. Finally, on the connected home consumer side, a big 20-nanometer play, as well as eMMC and SSDs to round out the storage side of the application, helping with the bandwidth needs of 4K. Okay, moving into mobile. The story here, quite honestly, is just still a continued spectacular content growth per phone on average.
What we've done here is tried to plot out the CAGRs across the various phone categories, what we're seeing, as well as the critical minimum densities, both on the memory as well as the storage side, to make what we're seeing and hearing in terms of the minimum densities to go out to market with. As you can see, at the high end that we all used to focus on, the fact of the matter is, these days, three gigabyte DRAM is a minimum. As a matter of fact, four gigabyte, even six gigabyte phones are coming to market and showing the fact that these mobile phones are true compute platforms these days. The other story here, though, is the sheer unit growth at the low end and mid end with a push in the memory and storage content to really make those phones usable.
That's driven by the displays, that's driven by video. A big push up here across all ends of the spectrum, showing how these phones really are true compute platforms driven by video applications and imaging. This segment here, specifically on the DRAM storage and emerging front, a lot happening. On the DRAM side, 20 nanometer, 8 gigabit components ramping hard now. 8 gigabit LP3 as well as 8 gigabit LP4. Obviously, a huge push here, specifically in power, to make sure that we have the lowest power consumption for our key partners. On the storage side, the proliferation of eMCPs, multi-chip packages, DRAM and NAND in the same package together, continues for form factor reasons to make sure that the memory of the storage is as close to the processor as possible.
Something worth talking about here, a number of our key partners in the mobile space showing huge interest in the emerging technologies that we've announced, as well as some of the technologies that Scott has alluded to. The big play here, looking for ways specifically to keep power consumption down, looking for faster forms of storage, as well as a memory technology that has persistence to it. You can understand the value on the mobile side, specifically on power consumption and what that means for battery life. A good portfolio here and something we're pretty proud of. Moving over to storage. Really here, across the entire SSD space, what you see is leading-edge NAND densities really transforming the space.
In the enterprise side, true enterprise SSDs to date have primarily been focused on just pure IOPS on the performance side, meaning the fastest possible random latencies on the PCIe side, as well as the reliability that you can get from SAS SSDs, really the traditional hard disk drive form factor, but placed in the form of an SSD. What's happening with the next generation systems, though, as those become enabled for SSDs, it's allowing the density to come up at the same time that the IOPS absolute performance is maintained. Across the cloud space, data center drives here, the play hasn't been so much IOPS, it's been pure density, okay? The scale-out play, what we call the servification of storage, getting SSDs inside of every server, inside of the data center to take over a big piece of the role of old traditional storage.
What it's doing is pushing on just density here more than anything. Really a rapid transition past 2 terabyte type drives. An appetite for significantly larger, using the old SATA form factor, good enough specifically for the performance needs across this wide dispersion of space. Again, just a very, very large push on density. Client side, as Mark Durcan indicated, we don't believe we've seen the tipping point here yet in SSD penetration across the client segment. Certain sub-segments have started to tip, notably commercial. The fact of the matter is that the world is waiting for cost-competitive 256 gigabyte and 512 gigabyte SSDs here, with penetration rates that are hovering, overall, in the client space around 25%. Just a huge amount of opportunity here. That gets us to 3D NAND, something that we see as a real game changer.
We are happy to announce today production availability and supply of a true TLC 48 gigabyte component, okay? That is a single monolithic piece of silicon optimized for SSDs. We're rounding out that portfolio with smaller densities that are beneficial across the board here for all kinds of applications. Think about 48 gigabyte component storage and what that means for cost competitive as well as high-density enterprise SSDs. We put our money where our mouth is. The fact of the matter, this technology is tooled up, as you saw from Mark's and Scott's slide, by fall of 2016. This will be exceeding our planar supply in bit terms. This is real technology. It's yielding well. We've got the production capacity behind it.
On top of that, what we're trying to show you is how we roll this out in terms of the end application space, the products our customers are calling for. These are products we're participating in today. We do view 3D NAND and the TLC portion of that as a real game changer here, helping with the cost and the density side in particular. Consumer SSD, starting in the second calendar quarter here, really helping with the cost profile, and drives. These are replacement drives for the end markets, as well as getting us out with high-performance drives to the end users. The client and OEM drives, the quarter after that, shipping directly to OEMs. A portfolio, again, you can see the density side here. One terabyte drives enabled specifically on single-sided M.2 form factor.
This has become the standard form factor for new drives shipping out of OEMs today, and something enabled very, very well by 3D NAND. As we've told you before, a good source for us here, shipping enterprise OEM components, strong uptake by our key enterprise partners here, specifically because of the cycling that Scott showed you, the reliability, as well as the density combined here with 3D NAND. CQ4 this year, coming out specifically with our own hyperscale SSD drives. Data center drives here optimized for density. We told you about the appetite specifically for two terabyte and beyond drives today, 3D NAND enabling drives up to the eight terabyte and beyond range. Something fairly powerful here. Then true enterprise high IOPS rollout CQ1 2017. Something we're pretty excited about across the board here with 3D NAND. 3D XPoint.
Back in August, you heard Micron and Intel together give you a few details of the technology. I'll try to put this in a big picture perspective. In technical terms here, in terms of the application space, what we're seeing here is a truly new revolutionary form of memory that is in between traditional memory, as implemented by DRAM, and traditional storage, as implemented by NAND and hard disk drives. Something that fills a key gap in between those two technologies. In density terms, think about rough densities this way. We've talked an awful lot about 1 gigabyte type DRAM densities that have been enabled with 20 nanometer. Conversely, I've mentioned the 48 gigabyte NAND component enabled with 3D NAND. This initial component will be a 16 gigabyte monolithic component. What you have is something that is right in between the two technologies.
It gives you some proxy for how to think about the rough density applicability. In terms of what that enables, it is byte addressable, which means it's closer to DRAM. It doesn't have these very long block write and erase times as NAND. It is asymmetric in terms of the reads and writes. It's a faster technology than NAND when it comes to read. It does have a longer write, though, compared to its read time overall. What that means is something that fits in very, very well between DRAM and NAND. What we do with that here across a range of applications is give something that can act as a very, very fast storage, as well as something that, in a number of ways, can act as a persistent form of memory.
The way the world's reacting to that, our key partners are seeing something that they believe across a number of applications will be additive to the entire space. Segments that generally fall into the category of non-consumption today. As Mark mentioned, in-memory databases, the kinds of densities they would like to enable, terabyte-type densities, multiple tens of terabytes, in-memory databases enabled with this technology. What we're doing with that, 2016, this is a year for us of technology development. As Scott mentioned, getting this technology sampled to our key partners, and working with them specifically on their ecosystems to make sure across the SoC, across the operating system, across the application software, this can work well. We see this rolling out in client for high-end gaming systems, specifically looking for the sheer random latency data center, specifically because of the higher speeds compared to NAND.
High-density storage and what that enables, a huge data center play after client. Like I mentioned, a huge amount of interest coming in specifically on the mobile side. These aren't easy-to-enable applications. Mobile will take a little bit longer to make sure the system is enabled well. Really here, thinking about the power consumption, and what that means for battery life, what it means for overall performance. Something we're very excited about here, a few busy years for us, and anxious to get this out to the end applications. Summing this all up, again, tried to give you some picture here of the opportunity, and we believe Micron is positioned very well, again, because of these three key pillars.
The memory technology, the portfolio of cost-competitive DRAM, of a game-changing 3D NAND, and an emerging memory space, as evidenced by 3D XPoint and some of the technologies in Scott's lab today. Packaging, it is just as critical, getting these put together in a form that actually matters. Subsystem execution, we're seeing it today with eMCPs, with NVDIMM, with HMC, with SSDs. The firmware, the controllers, the software to put this all together and make it work. Then finally, the ecosystem access, really making sure these difficult technologies, the difficult physics, once we get those working well, making sure we get those out to the systems that matter, and making sure the software is there, the operating system is ready to go, and the SoCs know how to talk to the new forms of memory. Pretty exciting space.
With that, I am going to go ahead and turn it over to Ernie.
Thanks. Good morning. I am going to try, in the next few slides, to sort of integrate a lot of what you have heard Mark and Scott and Brian speak about. We will start with an overview of the 2016 investment plans. This is the operating priority slide that Mark ended his presentation with. What we have done is sort of pegged for you what our current plans are with respect to CapEx investment in each of those areas. Extremely similar to what we showed you in August. We think that DRAM at this point is probably going to be at the higher end of the range, roughly a third or so of our aggregate CapEx.
If you look at non-volatile memory, which would include the facilities investments that we are making in Singapore, we are going to be about 50%, give or take, with technology and product enablement that covers a lot of the things that Scott spoke about, and the capital required to support that innovation, picking up the remainder. In aggregate, our overall CapEx guidance is unchanged. We still think the quote unquote, "net to Micron" is right around $5 billion, and that includes roughly $800 million, give or take, for the two facilities investments. The first being the build-out of Singapore Fab 10, and the second being the R&D fab that we are adding space to in Boise. Quick update on the Inotera transaction. The new news here is that we have finished the first round of the banking meetings. In Taiwan, we had about 150 Taiwanese bankers show up for that presentation.
It was the largest assemblage of bankers that had ever been remembered by anyone in the country's history. With the strong support of Formosa and Nanya for the transaction, we do feel good that we are getting indications from potential lead banks that are right-aligned with what we have talked about in terms of the terms for the overall loan. We will be back in Taiwan toward the end of this month after the Chinese New Year recess to finalize those things. We are also working on what we can do to allow ourselves to have the flexibility relative to that equity component of the transaction. The first being the build-out of Singapore Fab 10, and the second being the R&D fab that we are adding space to in Boise. Quick update on the Inotera transaction.
The first being the build-out of Singapore Fab 10, and the second being the R&D fab that we are adding space to in Boise. Quick update on the Inotera transaction. The new news here is that we've finished the first round of the banking meetings, in Taiwan. We had about 150 Taiwanese bankers show up for that presentation. It was the largest assemblage of bankers that had ever been remembered by anyone in the country's history, with the strong support of Formosa and Nanya for the transaction. We do feel good that we are getting indications from potential lead banks that are right aligned with what we've talked about in terms of the terms for the overall loan.
We will be back in Taiwan toward the end of this month after the Chinese New Year recess to finalize those things. We're also working on what we can do to allow ourselves to have the flexibility relative to that equity component of the transaction. We're pretty confident that if we decide and determine that taking a different approach to equity is the right way to go, that we're going to be able to sustain that and have the backing to make that happen for us. Obviously, that's a decision we don't have to make for quite some time. We can actually go very close to the projected closing, which is the July timeframe, before we have to make a determination about how we fund that roughly $1 billion or so of the transaction value.
This is a chart that Mark showed you in August, at the last Analyst Day, with respect to our bit growth. And with what you've heard today from Scott and from Brian, we feel fairly confident that we are on or slightly ahead of those plans in terms of the aggregate bit growth that the company expects to see. Call your attention to the fact that we told you back in August that we expected the first half of this year to be a period where we were making transitions both from planar to 3D, from larger geometries down to the 20 nanometer node, and we're coming through that period here. We're about to enter the period where we expect to see fairly significant bit growth coming out as we exit our fiscal year in 2016 and move into fiscal year 2017.
Clearly, with the 1X enablement that Scott talked about a little earlier today, we foresee that opportunity to continue that bit growth and actually improve it over the course of fiscal 2017. Although we don't foresee a change now with the expansion of Fab 10X in Singapore, we can be responsive to market needs. We fully expect that the NAND market is going to be such that we're going to build that out as we currently plan to. I would call your attention to the fact that there is some flexibility there that we have relative to that expansion capacity. Translating those into cost metrics, if you will.
The left-hand side of the chart is precisely what you saw at the August Analyst Day, with respect to both a 2-year CAGR for bit shipments, as well as an exit 2017 versus an exit 2015 bit profile. You can see on track with each one of these, what we've added for you is a perspective on the cost side. You can see both NAND and DRAM making substantial progress. The first thing your eye is probably drawn to on this chart is the fact that for DRAM, the 2-year CAGR and the exit doesn't look so much different. I think it's important to remember that when we're comparing DRAM particularly, we're comparing 90 nanometer to 110. Don't forget that as a result of the acquisition of Elpida and some of the related acquisition accounting, we had a very light depreciation load.
For us to be able to sustain these levels of cost reduction, essentially bringing the DRAM business over the course of this time back to a full depreciation load, if you will, shows some pretty extraordinary progress and very consistent with some of the cost per bit metrics that Scott was speaking of. He was talking more about the cash cost component piece of this, which is materially lower, so that the cost per bit improvement is materially better for cash cost for DRAM than the aggregate because of that depreciation load. You can see, in terms of an exit, we think we're going to be somewhere in the realm of 65%-75% lower on the NAND side.
We feel very good about the progress we're making on the cost side, and feel good about the fact that while bearing the increased depreciation load associated with essentially retooling these fabs that had no depreciation load, that we're still able to deliver on these cost per bit reductions. As exciting or perhaps more exciting, what we've done is sort of painted a picture here for FQ1s of each of the last few years about our bit share in the market. You can see as a result of the integration that we did with Elpida and the fact that we essentially put the roadmap on hold for a little bit while we did that integration, as well as making the tough call to do a very focused execution on 3D.
We have actually lost a little bit of bit share over the course of the last couple of years, and that has obviously impacted the top line. With the type of performance that you see outlined today, the bit growth that you see outlined today, we foresee that by the time we exit FQ 2017, that we will be back in the range of bit participation in the market that we have enjoyed over the course of the last few years. There is a real opportunity there at the top line, in addition to complementing it with the cost reductions that we see. This is obviously based on our understanding of the market and some of the forecasts that we've shown earlier today relative to projected bit growth, as well as how we see the competitive landscape shaping up.
A real opportunity here for Micron to move back to some of the ranges of bit market share that we have had previously. Finally, a quick update on our capital management framework. You can see on the bottom left with the integration of Inotera, we think the CapEx to sales is going to be up a little bit, about 30% or so. Don't forget, we're going to have full access to that cash flow. The Inotera transaction should generate somewhere around a net of that increased capital investment of $500 million-$600 million a year at today's market conditions. We still have a long-term objective of ensuring that our return of assets is greater than our cost of capital, and certainly on a last 12-months basis. That has been the case. No change to our targeted minimum cash balances.
If you do the math there, that equates to somewhere in the low $3 billion range in terms of a targeted amount. We could clearly run the company on a lot less if we needed to, but we have a very conservative perspective on that. That's where we'd like to be there. We continue to target GAAP debt to EBITDA, something less than 1.5 times. Again, if you look at the trailing 12-months, we're falling right within that range. With the acquisition of Inotera and depending on how that financing goes, there may be a period of time where we pop up above it. Clearly, part of the benefit of having a long-term target, it helps guide our thinking in terms of how we manage that over time. Very quick summary. We continue to invest in technology. That is our first priority.
I've often said that taking away any of the capital that we've allocated to these technology decisions and improvements that we're making is sort of the last thing that I would cut. We want to be flexible and remain returns-focused. Obviously, we're in the middle of some pretty challenging business conditions. With that priority on funding technology, we'll be thinking about the returns piece of that in the context of that. No change to our fundamental orientation. We are very pleased that the Inotera acquisition will drive significant incremental free cash flow. If you look at the last 12-months or so, net of what we think the ongoing CapEx is, that's probably $500 million-$600 million a year additional for the company that we will benefit from essentially completing that acquisition. Bit growth and cost reductions are going to enable margin expansion. It's been a while.
We've been pretty flat in terms of the bit-outs, but now is the time that we expect to reap the benefit of the investments that we have made as a company. It's a pretty exciting time ahead. Finally, no change to what is fundamentally a conservative capital management strategy, and we'll continue to orient ourselves in that way on a going-forward basis. With that, I'll turn it back over to Mark.
All right. Thanks, Ernie. Hopefully, your takeaway is that this team is executing, that we really have positioned the company for success, and that we have an opportunity here over the next couple of years to really drive some significant outperformance. Long term, we think the memory industry is investable. We think it's a market that continues to improve, that the dynamics in the industry continue to improve in terms of our ability to differentiate our products, deliver value-added systems. We think it's a market that continues to outgrow other pieces of the semiconductor industry and can grow more substantially, even in a market where potentially slower global economic growth slows unit growth. The underlying demand for memory per unit across a broad swath of market segments is such that we really think it's a good place to be. Operating performance is very good. This team is executing.
We're walking the walk, and I think we have all the resources in place. We're through our integration period with Elpida, and we think that we can continue to do this on a go-forward basis over the next couple of years. We've kind of outlined what we think is going to happen with supply. Our opportunity here over the next couple of years is to move back into those historical ranges as we outperform on the DRAM front and more substantially in 2017 on the NAND front as well. Really, very exciting time for us in terms of our ability to really move the needle relative to our deployed technology. We've got all the pieces in place to address our customer needs across a broad swath of end applications. We've got the broad product portfolio. We've got the system capability.
I've never been more excited about where we are relative to our ability to deliver the systems and subsystems in both Storage Business Unit and Mobile Business Unit in particular, but also in the other business units, to deliver the systems and subsystems functionality that our customers want. It's kind of coming together with this component roadmap and technology deployment that we talked about today. Finally, we think that you'll start to see that. We've been telling you for a number of quarters now that it's coming, and it's coming. We do have a lot of new things, a lot of new products, new densities, new interfaces, in both the DRAM and the NAND market. We've got a lot of segments and sub-segments, qualifications to work through. That's all going to happen as we move through the next couple of quarters.
We think that you'll be gratified with the performance of this company. With that, let me invite the speakers back up. We also, in the room here with us today, we have all the business unit leaders as well as some of the folks from finance, et cetera. If you ask us something really difficult or something where I think that they can add some added flavor, I'll certainly feel free to ask Mike Rayfield or Darren Thomas or Tom Eby or Jeff Bader to jump in as well on some of these questions. Let me open it up, and I'll kind of direct here the questioning, where we're going to take the questions, if that's okay. We'll start back here.
Yeah, everybody please wait for the microphones if you don't mind. One sec.
April, right here.
Hi, this is Srini from Summit Research. The main question that I have is on 3D XPoint. You guys did not give a bit shipment guidance for next year. What kind of bit shipment do we expect on 3D XPoint supply next year?
Brian?
Yeah, I can take that one. 2016, I think as we mentioned, I would treat this as a technology deployment year. A lot of samples going out to our key partners, getting this enabled in manufacturing and also getting it enabled in systems. We're not going to talk about specific bit shipment targets for 2016. Consider this a deployment and enabling year. Very bullish about the technology, but that's really the kind of timeframe to keep in mind.
As a quick follow-up, when you guys are saying that you'll have a relative margin expansion in second half of 2016, how much can we expect?
Yeah. Ernie, just before he jumps in, we're not predicting pricing for you.
Sure.
When we talk about margin expansion, it's relative to-
Costs
Our costs and also relative potentially to our competition. Go ahead, Ernie.
Yeah. I don't think there's a lot more to add to that.
I wanted to make sure I got that out.
Yeah. We aren't forecasting margin expansion. We can talk to you about cost reduction. We can combine that cost reduction with what we believe will be a better position, in the storage space with all that NAND output coming out. If we do our job well, it would certainly be there. It's difficult to predict in an industry where pricing moves around quite a bit, exactly what the margins will be.
Yeah. Assuming like 40%, 30% drop per year?
We can speculate all day long about what the pricing drop will be, but it doesn't speak to, for example, any value that we get from entering the SSD business and adding more value there. We don't do hypotheticals on margin.
Thank you so much.
All right, Mark. Okay, we got a mic over here, so go ahead. Then April up here next.
Hi, Karl Ackerman from Cowen and Company. Your closest peers who use charge trap are saying that 48-layer 3D NAND is not cost-effective versus planar in 2016. What's different about your 3D NAND that gives you 25% cost per bit improvement? I guess as a follow-up, what's your expectation of 3D NAND wafer starts exiting 2016? Because I think beyond the largest peer, most are expecting relatively little output of like 20 to 30K in 2016. Thanks.
Scott, you want to?
Sure. A couple of fundamental reasons for the reason our 3D NAND does cross over with planar technology. One is our ability to put the circuitry under the array and get a much smaller die size for a given bit density. The second one is our yield ramp being ahead of schedule, actually also goes to our ability to cross over faster with planar NAND technology. We believe we've built an efficient process technology base that gives us a solid cost position there also. I think we're very confident in just the cost benefit of our first generation of 3D NAND relative to planar.
Brian may want to talk more to the wafers or Mark also, we're also confident in the commitment we've made to the bit output increase on 3D NAND that we showed on a couple graphs and that Ernie described, really being substantial in the fall. I think you'll see a real focus from us on 3D NAND output and, specifically TLC 3D NAND output.
Let me jump in on the volume piece. We are in manufacturing today. We are now starting to run wafer volumes in the range that you just alluded to, and we're still in calendar Q1 of the year. We've showed you that we have a substantial bit ramp that happens in 2016 and even more coming in 2017. We're confident in our ability to deliver that, or we wouldn't be showing it to you. We have incremental clean room space we'll be moving into in the March timeframe that enables us then to drive based on ROIC, to where we want to get to. There is a real ability for us to drive this based on market conditions and a real belief that we have a very significant technology position that only gets better as we roll into our gen 2 technologies.
Did we cover everything that was in there? Okay. All right. Yeah.
Hi, thanks for taking my question. We've seen China's aspirations to enter the memory market. They have even talked about spending significant amount of money if they needed to open a fab. We also know they do not own any memory technology. The question is, what kind of relationship Micron could consider to work with China, or do you think it doesn't make sense to bring a new player in the memory market?
That's a topic we could probably spend all day on in this room. Let me just start with, we understand, I think everybody in the room understands that China would like to build a domestic semiconductor industry, that memory would be an important part of that. We have great relationships in China. We have very good customers from all the big manufacturers, down deep into the smaller industrial segments, et cetera. We have large operations there already today, with a significant test operation, starting up a joint venture in Xi'an for packaging, significant design firmware and software resources working to develop new products and technology within China. That relationship is pretty good. When we think about what we might do or might not do in China, we have to think about what does the market need.
Because at the end of the day, Micron needs to represent Micron's interests as well as what different geographies of the world might want to accomplish. Within that context, we have to think about what is likely to happen relative to supply and demand in the marketplace, and is new product really required, or greenfield fab's really required. We've told you today that we think DRAM demand may be just slightly above what could be accomplished with technology transitions, and that there's probably not a lot of new supply needed there. We've talked in terms of wafer starts. We've talked about NAND, potentially more capacity being needed over time. At the end of the day, we're going to do what makes sense for Micron shareholders.
We're going to listen to what opportunities might be out there, and we're going to think about what might make sense, but it has to make sense. Where we find ourselves today is that there are potential opportunities that are worth evaluating. The overall situation is such that we've got options in front of us with our own capacity that's already in development. Anything that we might contemplate has to really be very compelling, or it's probably not going to make sense.
Just as a follow-up, if the 3D XPoint benefits you talked about in the server market today, it seems like if 3D XPoint takes in the server market, it could be cannibalistic to your server DRAM market. Would you consider maybe converting your DRAM capacity to the XPoint at some point?
I can take that one, Mark.
Yeah, go ahead.
Great question, Monica. First of all, again, from an application standpoint, we're pretty bullish about 3D XPoint in a huge number of applications. What we're hearing from our key partners is that really for the problems they're designing for in the next generation systems, again, they don't see this as an either/or. They need vast amounts of very fast read and write DRAM scratch pad memory. They need vast amounts of solid-state dense storage, and they see 3D XPoint as something that goes in between those. Now, how we take a look at overall capacity, we'll navigate those decisions down the road, based on where the demands are, where we see the value-added products. Really, in terms of what DRAM's good for, what 3D XPoint's good for, and what 3D NAND's good for, will there be some cannibalization? At the highest end, it's always possible.
We just don't see it as a large concern.
Up here. Oh, okay, over here.
Thanks. Mark Newman from Bernstein. I wonder if you could talk a little bit about demand. Appreciate all the strong, compelling story on the cost declines. Also, in terms of content per box needs, due to performance needs, there's obviously going to be a lot of improved increased content per box, both DRAM and NAND. The missing part, I guess, I'd like to get some comments on is actually units, especially smartphones and PCs. There seems to be a lot of bearishness out there. I wonder if you could comment on where we are right now and if we're going to see some demand pick up a little bit this year.
On the general question, we believe that 2016 is much better than 2015 for PC demand. I think that's sort of the consensus of most people, that from a unit growth perspective, unless there's some sort of global meltdown, the PC business will be better than it has been. We see in the mobile space strong unit growth continuing in the future. Yes, there's more at the low end than there has been at the high end, as Brian pointed out, and as you guys are no doubt aware, that content per unit continues to grow significantly, even at the low end. We do see that there is unit growth in both those markets, both those big markets, on a go-forward basis. Well, maybe not unit growth on the PC front, but relative unit stability on the PC side. John.
Sorry, Harlan, I'll get to you next.
John with Credit Suisse. Thanks, Mark. Maybe a question for you and/or Scott. If you run the numbers, if you were to close half the margin gap between you and Samsung in your DRAM business, it'd be worth somewhere between $1 billion-$1.5 billion in free cash flow. If you close it all the way, it'd probably be worth about $2 billion-$3 billion of incremental free cash flow. Notwithstanding the benefits you're going to get as 20 ramps, can you help us understand more from a longer-term perspective how you structurally close that gap with the leaders on the DRAM side? Conversely, as you think about the 3D NAND technology you're bringing to market, should we just assume that's a significantly smaller gap between you and the leaders on the cost side than DRAM, or help me understand that.
The 20-nanometer ramp's going to be great, if we're going to be back here 18 months from now as we're talking about 1X or 1Z, it's not sustainable. That's a problem.
Yep. Hopefully we already convinced you that there's a pretty good path coming behind the 20 nanometer with the 1X. Scott, why don't you address the question generally, and then maybe I'll wrap up.
Okay
Any loose ends there.
Internally, when I talk about our direct focus on improving our DRAM competitive position, it absolutely contemplates what you're talking about. A very detailed recognition that, yes, 20 nanometers is going to improve our position substantially over this year. The game is really how fast do we get our 1X deployed, and do we really have a position with our 1Y and our 1Z to be in a very strong position relative to competition. Part of what we did when we rolled out our R&D strategy after the Elpida acquisition really was say we're going to go fund parallel development on multiple nodes in a way that we've never been able to do before.
That doesn't turn into results overnight as technology rolls out, it does say we're taking very serious what your question leads to, which is how do we close those gaps, and really improve that, ultimately, free cash flow in the future with more competitive technology rollouts. On the NAND side, I actually think we're in a very strong position. We have to prove it over this year as we roll out 3D NAND. I think based on the product portfolio that Brian said that is underpinned by the technology that we're putting out at a really competitive cost position, I think when we're standing here next year talking about this, I don't think we'll be talking about where are you on 3D NAND. I think we're going to have shown you.
Yeah. The other thing on NAND is, over the last couple of years, we clearly missed the boat on TLC enablement in all the non-consumer applications, and we're going to fix that. There's almost sort of a double whammy that comes as we go through that transition here. Harlan.
Thank you, Harlan. Harlan Sur with J.P. Morgan, thanks for hosting this event. To follow up on John's question, primarily with 3D, you gave us a really good view of your 3D technology relative to your planar 16 nanometer Gen 2 relative to Gen 1. Your competitors are rolling out 48 layer now, can you give us some idea, assuming you hit all of your targets, Gen 1, Gen 2, how are you going to stack up cost per bitwise relative to your other two or three competitors that are rolling out 48 layer?
Sure. I'll start anyway. I think, as I mentioned, our first generation was targeted in a way that already crosses over with planar NAND. I think from the details of your question, I think you probably know others have said that even their 48-tier technology is right on the edge of whether it crosses over. Some may or some may not. We can't predict that exactly. Based on our knowledge of where planar NAND cost structure really is, I think we're confident in where not only our 32 tier comes out relative to competition, but also the substantial benefit of our next node. Between those two, I think we'll be very competitive through the next year more.
Do you want to add anything on costs, Ernie?
No, other than I'd refer you back to those cost per bit charts where we talked about that exiting cost. I think you're looking at somewhere 75% or so lower than what we experienced as we exited 2015. Obviously, the Gen 1 is at a way station in between those two.
It's not really embedded in your question, but I think the other thing to keep your eye on in the NAND market is the relative value of the 3D NAND bits relative to the legacy planar bits that are in the marketplace today. That's not something we talked about at all today, but Scott did provide you evidence that these 3D NAND bits may be significantly advantaged relative to their performance and quality, and therefore, you might expect to see over time something additional that's not just cost-focused from this transition.
Going back to the profit differential in DRAM versus Samsung and others, can you talk about the trade-offs of the M&A that you've done? When you integrate Elpida and you have different tool sets, did that slow you down on '20 at all? Obviously you get the benefit of lower depreciation, but doesn't that kind of go away in the next couple of years?
Yeah, let me take that one. Absolutely, Joe. Without a doubt, there is a price to be paid or a price that has been paid for the way we've grown. This company is twice the size it was four years ago. It's not all bad. There is a price to be paid when you take disparate tool sets and you have to reintegrate them, and you have to integrate roadmaps and redevelop processes to work on multiple tool sets, et cetera. That penalty dissipates over time as the team becomes more functionally integrated and as the tool sets get replaced out through subsequent generations. We feel like we're at now really on the cusp, three years on, to being able to really start reaping those benefits. We've got a commonly defined technology node that we're rolling out.
We're on our second generation of some level of tool replacement, that's all beneficial on a go-forward basis. The depreciation story, yes, it's true that when we put a bunch of written-down assets onto our books, we had initially a very depressed depreciation as a percentage of our total manufacturing cost. As we bring new tools online, that depreciation level as a percent of total cost tends to grow for a while, then that'll saturate, and it'll be more consistent. We're working through that process as well. All those cost numbers that Ernie showed you embed that effect. The actual cash cost reduction that Micron's going through actually exceeds the numbers that we're showing you on a go-forward basis.
Thank you.
Over here. Yeah, fine.
Thank you. Question for Ernie. You had a chart talking about your bit, the market share rebounding, especially on DRAM, what gives you confidence that the competitors are going to give up on the share?
We have taken a look at the comments they've made, what they've said about their capital investments and their technology migration. Obviously there's probably 20 variables that go into the chart that I showed. At least based on everything we know, we think that that is a reasonable goal for us to drive towards over the course of the next year or so.
Let me just add. Our goal here is to deploy advanced technology and to make sure that we're running effective and efficient manufacturing operations. What Ernie showed you is what you reasonably might extrapolate. We don't know what our competition's going to do. If they want to mess up the market, I guess they can. That might happen, but that's not what we would predict based on what would seem to be a prudent course of business and what they've said. Additionally, what I would say is we're not out there, at least in the DRAM business, actively adding wafers to the marketplace. What we're doing is we're just deploying the same technology that our competition's already running.
Fair. One question for Brian.
Just looking at the big picture, structural changes taking place, more of a traditional enterprises tapping into a hybrid cloud approach.
Yeah.
The public cloud guys increasingly trying to disaggregate compute from storage. Given what I just said, in that context, how should we think about the XPoint? What are your key assumptions for end market demand, especially on structural changes and given what's happening to the cloud, what gives you the confidence that there will be this much growth for XPoint? What if the public cloud guys are actually successful with disaggregating compute from storage?
Sure. Great question. What gives us confidence there is, there are, to your point, a lot of trends underway, public cloud, private cloud. Putting 3D XPoint aside momentarily, I think one of the more interesting things we're seeing is that if you stepped back two years ago and looked at the predictions about the cloud space, there was a sense that the client compute platforms, the mobile phones, the end terminals were actually heading towards kind of the old VAX days of being a dumb terminal with all the hardcore compute happening inside of the cloud. Interestingly, that's actually somewhat been disproven where the content creation, the need for battery life, the need for real-time computing, and video and image happening at the client level has driven needs that a couple of years ago, two or three years ago, weren't predicted.
As we think about 3D XPoint, and the power that it enables, not just in the data center space, but really at the client level, specifically with persistent forms of memory as well as a much, much faster form of storage, I'd say some of the, frankly, the biggest interest to us is coming in on the mobile side. We're pretty enthused about it.
Thank you.
Go ahead, April.
Hi, this is Tanya Sell. You've given a good vision for the near term, the next couple of years. I'm kind of curious, you're going to spend 20%-30% of CapEx on things like firmware and all the pieces, putting it together, subsystems and all of that. Beyond, call it two years, how do you envision your business model evolving? Because that's a little bit different business. You're going to maybe need to invest in more headcount. How do you see that playing out, or how do you guys look in a few years?
There's a lot of activity in that bottom bucket of CapEx. Yes, it's investments in controller hardware and firmware and software, it's advanced packaging and other emerging memories, et cetera. A lot of stuff goes into that. The long-term vision of what the memory business looks like is a business where customers buy memory systems. Hopefully we've given you some sense as to why that is. It's because the technologies are evolving, and in some cases, they need to be married with additional functionality that's not what you would historically build into a single component, it incorporates control functionality and closer integration with the end system. Also because in the future, really to deliver the optimal functionality, you have to marry multiple memory types. That's what we view our business is becoming. It's becoming a memory subsystem and a memory system business.
In order to deliver that in the future, and today, we have to continue to make those investments. There is a return, and we actually think that it's well worth investing in because in many ways, there's a better ROI on those types of investments than in just a pure hardware business.
I can imagine. Thank you.
Over here.
Hi. Thanks for taking my question. I'm C.J. Muse, Evercore ISI. Ernie, question for you on the cost side. You laid out a 15%-25% cost down, including in Inotera over the next two fiscal years. Curious if you could share what you're thinking for fiscal 2016 and then fiscal 2017. Is more of it weighted to the second year or the first year? If you can also elaborate on what gets us to the low end versus the high end.
Sure. I think consistent with the bit growth that preceded that cost reduction chart, you would expect to see more of that start to blend in the back half of fiscal 2016, but certainly fully realized in fiscal 2017. If you put on top of that the timing of the Inotera transaction, which is going to be closing, if all goes well, in the July timeframe, certainly the full impact won't be realized until fiscal 2017. I think it's reasonable to think about things in that way.
The low to high end of the range, obviously, that depends on yield execution.
That's right.
On the NAND side, relative to exactly what investments we decide to make at the end of the day. This will be return based. Over here.
Thanks for taking my question. Doug Freedman, Sterne Agee CRT. I want to build a little bit on some of your prior answers. In talking about your tool set not being similar across your fab footprint, is your CapEx budget over the next couple of years sufficient enough that you will end up ultimately with copy-exact tool sets across your factories? When does that happen? When that happens, will it enable you to ramp new technologies faster at a lower cost? Is there an R&D savings that we might ultimately end up seeing in the next couple of years?
First of all, on the NAND side, we're already there because the NAND technology has all come out of our R&D fab, and it's all being deployed in Micron-only historical fabs. That activity is already there. Yes, it helps with R&D spend. It helps with technology rollout, and we can see that effect in our business on the NAND side. On the DRAM side, you never get there exactly unless you're prepared to go pull a bunch of stuff out and put a bunch of stuff in, and that's wasteful. Over time, it will become more and more similar, and there will be less and less discrepancy fab to fab. As I said, going from 30 to 25 to 20, we're already now on our second hop, at least in a couple of the DRAM fabs.
We go to 110, we will continue to blend that out, and we'll be largely homogeneous at that node and beyond. This never gets perfect, though.
Okay. If I could, for moving on to another side of your business, the new 3D XPoint coming up. You talked about it being of interest to the mobile market. How should we expect you and Intel, the two providers of the technology to the market, to segment the market? Are there segments where we should expect Micron to be dominant and other segments where Intel will be dominant? How should we envision that in the future?
Yeah, great question, Doug. As we've told you before that, first of all, the JDP between Micron and Intel is related to the development of the core technology itself. Both parties are completely independent in terms of go-to-market strategy. Micron has end segment focus. It's been very, very good to us over the years. Mobile is a huge piece of that. Certainly data center as well, data center storage, compute applications. Obviously Intel has their own particular direction. I won't speculate in terms of how they roll out the technology. Both sides are independent on the go-to-market side. What I do know is that, to put it bluntly, we're overwhelmed with the interest coming in
From our end segment providers, they see the value there across mobile, enterprise, and even client.
Oh, go ahead.
Thanks very much. Mark Delaney with Goldman Sachs. Thank you for the presentations. Question on the 1X nanometer DRAM rollout. How much of your mix should we expect to be on 1X nanometer DRAM as you exit calendar 2016 and into 2017? Are you taking 25-nanometer wafers and converting them right to 16, or is it going to be a complete transition to 20 and then on to 16 after that?
In our fiscal 2016, you won't see substantial output on our 1X node we're deploying in several months. In 2017, it'll be substantial, and I think there was a picture of the bit output ramping on one of those graphs, if you pull that one back up on the 1X, really coming up in the second half of 2017. There'll definitely be bit output earlier than that, but when it starts becoming significant to our business is.
Fiscal 2017.
Fiscal. All my statements here are fiscal. Second half of your question was?
25 to 1X.
Depending on the facilities that we put the 1X nanometer technology into, your question's answered in a little bit different ways. In some cases, we'll be converting 25-nanometer technology. Certainly, if we haven't converted that yet to 20-nanometer, then we would go straight to 1X. In other facilities that have fully ramped on 20-nanometer, then we would be deploying 1X into 20-nanometer.
Vijay.
Vijay. This is Vijay from Mizuho. Just on the DRAM side, when you look at your 20 nanometer coming on, is it all DDR4 that's coming out now? Also on the NAND side, you talked about gen 2 3D NAND. Is that 48 layer or 64 layer? Thanks.
On our generations other than gen 1, we're not being specific on the number of tiers yet. Part of that is because, as I mentioned before, all 32s aren't created equal, all 48s aren't created equal, so we're really more focused on the bit density improvements we get between nodes. There's about an 88% increase in gigabits per area on our generation 2 relative to our generation 1. That's a combination of things, number of tiers, design efficiency, and a variety of other things.
I can hit the.
Thanks.
I think the first part of your question was related to 20 nanometer DRAM and the market penetration. That is, for us, on 20 nanometer, a mixture of DDR3 and DDR4 today, consistent with market needs, and on the mobile-
Elpida. Yeah. Sorry.
On the mobile side, LP3 and LP4. As we exit 2016 calendar year, the majority of the bits will be on DDR4 and LP4.
All right. Thanks.
Bill?
A couple of questions. First one, relative to the rational nature of the industry, when you made the Elpida acquisition, there was a hope and a belief that the industry would be more rational, and clearly it has. The question is, how would you characterize the level of rationality relative to what you had expected at the time that you began that transaction?
Yeah. As I look at the market today, I don't see a lot of what I would classify as aberrant behavior or destructive behavior in the marketplace. I think given what played out relative to demand in 2015, probably most market participants wish that as many DRAM wafers hadn't been added as were added. I think that actually, as I look at what's going on in the marketplace today, I'm actually quite enthusiastic about the memory industry on a go-forward basis and our ability to drive sustainable good margin in the business.
Thank you. The second question is relative to demand, and this is somewhat Micron-related, but somewhat global economic-related. We read in the newspapers that the economy may be heading to worldwide depression forever. Tell us, what are you seeing from a demand standpoint since you reach all aspects of the world?
Well, I think maybe some of the other folks want to comment here. I think generally speaking, things are ticking along as we expected. As we look at what's going on in some of the larger market segments, as we look at what's going on in the industrial, medical, automotive segment, there's continued growth there. Maybe it's not quite as robust as we thought it would be, but it's ticking along fine. We don't see any catastrophe coming. We grow, as you pointed out in your question, Bill, we grow with global GDP because we play in all those markets. We are levered to that. We're not seeing anything unusual in terms of dramatic, steep roll-offs in our end demand in really any segment.
Thank you for not predicting Armageddon. Hi. Karl Ackerman, Cowen and Company. Just to follow up. Your stock is currently trading slightly below your reported tangible book value.
I think Elpida wrote down over $2 billion of PP&E in bankruptcy, and you also wrote down quite a bit after the acquisition of Elpida. Just curious if you could help us understand what you think your true book value is today, and how much Inotera would add, assuming the deal closes in 2016.
Sure. If you think about the write-downs that occurred, but the fact of the matter is some time has elapsed as well. I think the total impact of that Elpida acquisition is probably net null or maybe slightly positive to where we are today. Clearly, if you add everything that was written down and assume that none of it got further depreciated, you'd be up another $2 or $3 billion, give or take. Relative to Inotera, we think that that, from a book value perspective, is going to increment by roughly the same amount as we bring it fully on board.
Thanks. A question on the 3D NAND. The decision to put the logic underneath versus on the die, does that give you a better usable area of memory on the die versus competitors? Is there a metric you can provide there? Was that a decision at all related with your choice using floating gate versus charge trap? Are those two related or separate decisions?
Those two decisions are definitely coupled in not saying that it's impossible the other way, but it definitely is an enabler of being able to put the CMOS under the array. The combination of those two is a powerful thing that drives cost reduction. I mentioned on that slide, for us specifically, it provided at least a 20% cost benefit to us if we'd have put it outside the array. The question's really complicated to answer and maybe too speculative for me to say exactly what it would've done relative to our competition, because there's a lot of different design efficiency combinations that go into ultimately how you make a NAND die. Definitely at least a 20% benefit. You can extrapolate that to competition would get at least the same if they were able to put the CMOS under the array or more.
For sure, a difference in the usable area of the die of memory versus competitors.
Yes. There's a measurable difference in the number of die per wafer for a given density, obviously, because of the design efficiency.
Sid, if you're able.
Thanks. With regards to your own CapEx this year, I may be splitting hairs here, but sounds like you uptake that a little bit. In contrast, some of your competitors are downtaking that given the market conditions. Should we expect that to be the minimal for your technology deployment you talked about, or is there some wiggle room if things continue to be weak?
Part of what we've talked about pretty consistently throughout the day is this is a year where we are closing the gap, and as a result of closing the gap, we're fairly committed to that level of DRAM CapEx under all but the most extraordinary conditions. I would point out that we're at the upper end of the range that we provided you in August. We haven't gone outside the range and don't expect to either. One last one here.
We've got another one over here too after that.
Two last ones.
Okay. Right here, Melissa. Thanks.
You talked about how a lot of what you're doing this year operationally is enabling new products for 2017. Can you sort of maybe timeline or generalize how all that comes together and maybe when we see a better mix of business from your new products as you work out? You talked about new product starts, but you didn't talk about how the volumes ramp and where we could see maybe a sweet spot for that in 2017.
Let me make a general comment, and if we need more detail, I'll kick it over to Brian. Really talking about two different things, right? As we move through 2016, we've got a lot of product transitions that come with the move from 30 and 25 nanometer production to 20 nanometer production. Almost the whole product portfolio is going to roll over into new products, so it'll have to get qualified and start shipping to customers. That'll happen as we move through 2016, because you see the volume coming up very quickly now in the back half of fiscal 2016. Relative to NAND, a lot of those new product transitions are tied to the 3D NAND transition. That happens also in 2016, but it's a much bigger story in 2017.
We feel very optimistic about our ability to deliver those NAND transitions, and we feel very good about the way that roadmap for the storage solutions is now very well aligned with the technology transition. That'll be a later in 2016 and into 2017 story as opposed to DRAM, which is going to start happening in real time over the next couple of quarters. Okay, we had one over here? Yeah.
Your New Memory A and New Memory B for DRAM and NAND spaces specifically. Does New Memory A and DRAM come in after the next three generations of DRAM because you noted 1X, 1Y, and 1Z, or does it intersect simultaneously? Second, are you developing these new memories on your own, or will the IP be jointly shared with Intel?
On the first question, I think just like when we talk about our 3D XPoint technology, it comes in over time, and we're deploying it now and enabling it. I think New Memory A, for example, is a technology that we'll deploy over time. It won't wait until after the 1Z DRAM technology node. In fact, we're pushing on it as hard as we can with some strong partners. That's a Micron-specific technology. On the other end, there is, again, opportunities and the timeline will grow, but we're focused on identifying, in that case, maybe even a harder technology challenge, which is getting to a cost per bit position that's substantially better than NAND.
The NAND roadmap moving as fast as it is, in terms of reducing the cost per bit, is going to be maybe more challenging to actually get something out in front of it and hit it. We think we have a technology identified, but it also is dependent on how fast NAND continues to move. That one's a little harder to answer. It could be toward the end of our NAND scaling roadmap, or may pull in depending on NAND scaling.
It's Micron specific, or it's joint with Intel?
Micron specific.
Ivan?
Thanks, everyone, for coming today. I think we're going to wrap up there now.
All right. Thank you all.