[Foreign language] Welcome to TSMC first quarter 2017 earnings conference and conference call. This is Elizabeth Sun, TSMC Senior Director of Corporate Communications and your host for today. Today's event is webcast live through TSMC's website at www.tsmc.com. If you are joining us via the conference call, your dialing lines are in listen-only mode. As this conference is being viewed by investors around the world, we will conduct the event in English only. The format for today's event will be as follows: first, TSMC Senior Vice President and CFO Miss Lora Ho will summarize our operations in the first quarter of 2017, followed by our guidance for the second quarter of 2017. Afterwards, TSMC's two Presidents and co-CEOs, Dr. Mark Liu and Dr. C.C.
Wei and Miss Ho will jointly provide our key messages. We will open the floor for questions and answers. For those participants on the call, if you do not yet have a copy of the press release, you may download it from TSMC's website at www.tsmc.com. Please also download the summary slides in relation to today's earnings conference presentation. As usual, I would like to remind everybody that today's discussions may contain forward-looking statements and are subject to significant risks and uncertainties, which could cause actual results to differ materially from those contained in the forward-looking statements. Please refer to the safe harbor notice that appears on our press release. Now I would like to turn the podium to TSMC's CFO, Miss Lora Ho, for the summary of operations and current quarter guidance.
Thank you, Elizabeth. Good afternoon, everyone. Welcome to join us today. My presentation, as usual, will start with financial highlights for the first quarter, followed by the guidance of the second quarter. First quarter revenue decreased 10.8% sequentially, but increased 14.9% year-over-year. The sequential decline reflecting mobile product seasonality and about 2% appreciation in TWD against the USD. When we give our guidance in the first quarter, we were assuming 1 USD to TWD 32. However, the actual average exchange rate was 31.16, which reduced our revenue by about TWD 6 billion. Without the TWD appreciation, our first quarter revenue would have been about TWD 240 billion, exceeding the high end of our guidance. First quarter gross margin was 61.9%, slightly lower than the fourth quarter 2016, mainly due to lower level capacity utilization and an unfavorable foreign exchange rate, partly balanced by continued cost improvement. Operating expense ratio rose to 11.1% as R&D as a percentage of revenue increased by 70 basis points. Operating margin decreased 1.1 percentage point sequentially to 40.8% in the first quarter. Overall, our first quarter EPS was TWD 3.38, and ROE was 24.6%. Now let's take a look at wafer revenue contribution by application. During the first quarter, consumer and computer increased 30% and 1% respectively, while communication and industrial standards decreased 18% and 5% respectively due to mobile product seasonality. Now let's take a look at revenue by technology. Combined revenue from 16 and 20 nanometer was 31% of total wafer revenue in the first quarter, while 28 nanometer represented 25% of total wafer revenue. Advanced technologies, which we define as 28 nanometer and below, accounted for 56% of total wafer revenue in the first quarter. Moving on to the balance sheet, we ended the first quarter with cash and marketable security of TWD 659 billion, an increase of TWD 27 billion from the fourth quarter. On the liability side, current liabilities slightly increased by TWD 3 billion. On financial ratios, accounts receivable turnover days increased two days to 47 days, while days of inventory increased three days to 44 days. Now let me make a few comments on cash flow and CapEx. During the first quarter, we generated about TWD 161 billion cash from operations and spent TWD 103 billion in capital expenditure.
As a result, we generated free cash flow of TWD 58 billion. We also repaid TWD 10 billion corporate bonds and net purchase of about TWD 7 billion fixed income securities. Overall cash balance increased by TWD 23 billion to reach TWD 565 billion at the end of the first quarter. In the US dollar terms, our first quarter capital expenditure was $3.3 billion. Full-year capital budget remain at about $10 billion. I have just finished the financial summary for the first quarter. Now let me turn into the second quarter guidance. We expect second quarter demand will be weaker than the first quarter due to supply chain inventory management during the second quarter and mobile product seasonality.
Based on our current business outlook and exchange rate assumption of one US dollar to TWD 30.5, we expect second quarter revenue to be between TWD 213 billion and TWD 216 billion, which represents 8%-9% sequential decline. Gross profit margin to be between 50.5% and 52.5%, and operating margin to be between 39% and 41%. The margin guidance reflect a higher utilization level for work-in-process building to support a strong 10-nanometer shipment in the third quarter. In the second quarter, we will again need to accrue the 10% tax on the undistributed retained earnings. Our second quarter tax rate will be about 23%. The tax rate will fall back to 10%-11% level in the third and fourth quarter, and the full year tax rate will be between 13% and 14%.
This concludes my remark, I will turn it over to Mark for his comments.
Good afternoon. Let me start from my message on the near-term demand and supply chain inventory. All our shipments in US dollars, allow me to just use the US dollar to describe our demand. We just concluded our first quarter revenue to be above our January guidance in US dollars. It gives us a -9% quarter-to-quarter in US dollars, which is about +22% year-to-year growth. This quarter-to-quarter change in 1Q 2017 was due to the seasonality of our major smartphone customer and a slower smartphone demand in China. We now forecast our second quarter revenue to decline about 6% quarter-to-quarter in US dollars. Year-over-year, this is an increase of about 3% year-to-year in US dollars. Together with the 1Q revenue, our first half 2017 revenue would grow about 12% year-over-year in US dollars.
This is slightly higher than the 10% increase year-to-year we forecasted in our January investor conference. We now estimate fabless DOI is still high, high above seasonal exiting first quarter 2017. Our second quarter revenue guidance reflects a quite severe inventory adjustment by our customers, particularly in smartphone and PC markets. The overall end market smartphone demand appears stable in second quarter 2017. We estimate the fabless DOI should approach seasonal level at the end of second quarter 2017, and demand for our products will be poised for a strong growth in the third quarter. Given the very strong demand in the second half last year, we maintain our second half 2017, this year, growth rate estimate of 5% year-over-year. Our full 2017 growth rate target remains to be 5%-10% in US dollars as we previously stated.
On the forecast of overall semiconductor market growth rate
Compared to three months ago, we raised it to 7% from 4% due to a stronger memory market. Semiconductor, excluding memory market growth rate, remains at 4% this year. We also revised foundry revenue growth to 5% from 7% due to this elevated inventory in the supply chain. Our 5%-10% full-year growth forecast shows we will continue to increase our market share this year. Thank you. I'll hand the mic to C.C. Wei. I'm sorry, Lora.
As the foreign exchange rate is volatile this year, it's almost very difficult to predict. I will make some comments on foreign exchange rate and the impact to our revenue and the profitability. As you may all know, NT dollars is the reporting currency for all our financial statements. Due to the fact that nearly 100% of TSMC's revenue is denominated in US dollars, and about 75% of our cost of goods sold, and about 70% of our operating expenses are based in NT dollars. Therefore, the fluctuation in exchange rate between US dollars and NT dollars will have a sizable impact to our reported revenue and the gross margins. The sensitivity of revenue to US dollar, NT dollar exchange rate is nearly 100%. That is, if every 1% appreciation of NT dollars to US dollar will reduce our reported revenue by about 1%.
The sensitivity of both our gross margin and operating margin to the same 1% exchange rate change is about 40 basis points. That is, if NT appreciates 1% against US dollars, our gross margin and operating margin will both come down by about 40 basis points. Compared with the first quarter guidance made in January 12th, the NT dollar has appreciated by an average of about 2.6% sequentially, which negatively impacts our first quarter revenue by about 2.6% and our gross profit margin about 100 basis points, gross margin and operating margin each. For 2017, the second quarter, we forecast the average NT dollars will further appreciate another 2.1% sequentially, which will negatively impact our second quarter revenue by 2.1% and reduce our gross margin and operating margin by about 85 basis points.
Should exchange rate have stayed at a first quarter 2016 level, which was 31.77 in average, our first quarter 2017 revenue would have been 1.9% better than the actual number reported here, and our second quarter revenue would have been 4% better than we just guided. Thank you. I will turn it to C.C. Wei.
Thank you, Lora. Good afternoon, ladies and gentlemen. Let me start with 10-nanometer ramp status. We have passed the reliability qualification on internal technology qualification vehicle and have also passed 500-hour reliability qualification on several customers' products. N10 has been transferred from R&D to operation in both Fab 12 in Hsinchu and Fab 15 in Taichung and is ready for high volume production. Although N10 technology is very challenging, the yield learning progression has been the fastest as compared to the previous nodes, such as 20-nanometer and 16-nanometer. Our current N10 yield progress is slightly ahead of schedule. The ramp of N10 will be very fast in the second half of this year. We expect the 10-nanometers will contribute about 10% of our wafer revenue this year. Now let's move to N7 and N7+.
TSMC N7 will enter risk production in the second quarter of this year. We have more than 30 customers actively engaged in N7, and we expect about 15 tape-outs in this year with volume production in 2018. In just one year after our launch of N7, we plan to introduce N7+ in 2018. N7+ will leverage EUV technology for a few critical layers to serve more immersion layers. In addition to process simplification, our N7+ provides better transistor performance by about 10% and reduces the chip size by up to 10% when compared with the N7. High volume production of N7+ is expected in second half 2018. I'm sorry, in second half of 2019. Right now, our focus on EUV include power source stability, pellicle for EUV mask, and stability of the photoresist.
We continue to work with ASML to improve the tool productivity so that it can be ready for mass production on schedule. Now N5. We have been working with major customers to define five-nanometer specs and to develop technology to support customers at risk production scheduling second quarter 2019, with volume ramp in 2020. Functional SRAM in our test vehicle has already been established. We plan to use more layers of EUV in N5 as compared to N7+. In addition to those leading-edge technology, we also continue to improve N16 and 28-nanometer, which are in mass production for many years. Now let me talk about the N12.
After we have developed 16 nanometer technologies from 16FF to 16FF+ and then to 16FFC, we further extend this technology to 12 nanometer, which will have about 10% better performance at the same total power, or 25% lower power consumption at the same speed. About 8 to 10 smaller die size compared with our 16FFC. Although 12 nanometer will have smaller metal pitches as compared to 16FFC, we have worked with IP infrastructure partner to build a complete IP support to make sure our customers' 16 nanometer product can be successfully ported to 12 nanometer with minimal effort. We anticipate the completion of 12 nanometers development by middle of this year. So far, more than 10 customer has actively engaged, with seven tape-outs being planned in 2017. The major applications will be mid to low-end smartphones.
With 12 nanometer having better cost structure and better performance than our already competitive 16 nanometer, we expect to maintain a high market segment share in this N16, N12 node. Now 22ULP. 22ULP is a half node of our 28 nanometer technology. We developed this technology to address the market segment where low operating voltage are required. Applications in IoT, image signal processing, GPS, Wi-Fi, and 5G millimeter wave are examples that can find good use of this technology. Compared to our 28HPC+, our 22ULP offers 15% performance improvement or 35% power reduction. It also reduces the die size by up to 10%. In addition, the RF performance, the cutoff frequency or the maximum frequency, it also improved to 400 and 370 gigahertz, which is useful in 5G application. We expect to begin 22ULP volume production in 2018.
With the enhanced features carried by this technology, we expect to maintain a high market share in this 28, 22 technology family. Now let me talk about the specialty technology. Especially this time, I will cover ultra-low power, which is very important for all mobile products. We have offered the industry's most comprehensive ultra-low power technology portfolio, including 55 nanometer, 40 nanometer, 28 nanometer, 16FFC, and now 22ULP and 12ULP. All these technologies are targeting IoT, mid to end smartphone, GPS, Bluetooth, and other applications. Depending on each application's performance and power requirement, the customer can choose their optimized solution from our portfolio. 55 nanometer and 40 nanometer ULP in high volume production since early 2016. So far, there have been more than 35 products running in the line. In while, 28ULP, 16FFC are also in high volume production.
In addition to those technologies, we have also developed near threshold voltage technology in 40 nanometer. This technology can achieve power consumption below 10 microampere per megahertz active power, which is 5 to 10 times lower than industry's best product for today. We are working on the design enablement support for our 14 nanometer near threshold voltage, which will be completed by third quarter this year. Now let me update on InFO. First, we expect InFO revenue in 2017 will be about $500 million. Now we are engaging with multiple customers to develop next generation InFO technology for smartphone application for their 2018, 2019 models. We are also developing various InFO technologies to extend the application into high performance computing area, such as InFO on substrate, we call it InFO-oS, and InFO with memory on substrate, InFO-MS.
This technology will be ready by third quarter this year or fourth quarter next year. That's all my update. Thank you for your attention. Mark?
Okay, let me share with you the last bullet. AI, artificial intelligence, and the ubiquitous computing. This is to share with you the recent development in the semiconductor industry and how TSMC position ourselves to ride on this trend. We are in a new era where billions of devices are connected at all times, and computing takes place at any time and any place. This is what we refer to as ubiquitous computing. For smartphone, for example, in addition to the smartphone unit growth, more intelligent features such as voice, image recognition, and AI for decision-making will further increase its computing power and silicon content. Given the vast established subscription base of smartphone today, it is the best launch pad for new consumer hardware and software innovation.
For AI in high performance computing, let me first define the HPC, high performance computing, as the semiconductor used in data centers, servers, networking, storage, and gaming. Artificial intelligence application and services and 5G infrastructures are the major driving force with leading-edge technology behind this HPC growth. For example, more than half of our customer product tape outs with our seven nanometer today we see are HPC products. Recently, we are also very encouraged by the support of a major AI service provider on Arm-based processor for data center in Open Compute Project Summit this year. Accelerators used in data center are also increasing, adopting the GPUs, FPGAs, and ASICs. We expect HPC to become our major growth engine from 2020. The trend of ubiquitous AI also shows up in many IoT and consumer devices, such as robot, drone, surveillance devices, smart TV, and set-top box.
Ubiquitous AI will also be widely used in the fast-developing autonomous car market. All these require very intensive, localized parallel computing capability, which drive up the silicon content. At TSMC, we work with innovators around the world. TSMC's long-term growth engine ride on this industry trend, from mobile computing to ubiquitous computing. The proliferation of AI demands insatiable computing capability from semiconductors. As C.C. just talked about our technology updates, we are developing various technologies and innovation platforms to satisfy this industry trend. Thank you for your attention.
This concludes our prepared statements. Before we begin in the Q&A session, I would like to remind everyone to limit your questions to two at a time to allow all participants an opportunity to ask questions. Questions will be taken both from the floor and also from the call. Should you wish to raise your questions in Chinese, I will translate that to English before our management answers your question. For those of you on the call, if you would like to ask a question, please press the star then one on your telephone keypad now. Questions will be taken in the order in which they were received. If at any time you would like to remove yourself from the questioning queue, please press the pound or the hash key. Now, let's begin the Q&A session. All right.
First one, the first questions will be coming from Credit Suisse, Randy Abrams.
Yes. Thank you. First question I wanted to ask about 28 nanometer, where you talked about still gaining a bit of market share. There's a couple rival options. Intel announced 22 nanometer using the FinFET that they have on 22. Then also the Samsung, GlobalFoundries have the FDSOI. For China, they haven't been a presence, but starting to ramp up. I'm curious how you're viewing these other options and the driver of confidence on the 28 nanometer relative to these.
Well, Randy, we continue to improve our technology, so we are very competitive. As I said, we developed 28 nanometer from time up to 28HPC+. Now we have a 22ULP, so we are still confident we can maintain a very high market segment share on this technology node. I don't want to comment on our competitors' progress.
The second question, I wanted to ask about the margins. You've actually done a great job weathering two quarters in a row, where margin's still above 50% and sales are down high single digit two quarters in a row and high margin. If you could take a forward look towards second half where we ramp up 10-nanometer and at these exchange rates, and since you're building WIP in second quarter, how you expect the margin to trend, if there's still some leverage in the model as we go into peak season in second half?
Randy, we intend to keep structural profitability at close to 50% margin. As you know, we are ramping 10-nanometer significantly starting from the second half of this year. I remember I said last time, there will be some dilution from 10-nanometer. In terms of the magnitude, our estimation is for the second half, you will have 2-3 percentage point dilution to a corporate level gross margin.
Do you expect offsetting that, any incremental leverage from the rest of the business to potentially offset the 10-nanometer dilution?
Usually at the very beginning stage of any technology ramp, it's difficult to find offset immediately, but when time goes by, and we can offset gradually as the 10-nanometer margin will also continue to improve.
Thank you.
All right. Next question will be coming from Deutsche Bank, Michael Chou.
C.C., seems that you mentioned before there will be only one 7-nanometer EUV customer. Are you seeing more customer for 7-nanometer EUV?
Yes. As I just reported, we are engaged with many customers on 7 and also 7+. That answer your question?
I mean, for N7+, right? Management mentioned before, only one customer for EUV in 7-nanometer. Now are you seeing how many customer you have?
Many.
Okay. The next question, sir. Management mentioned log scale comparison versus Intel, I think the 2014, right? Since Intel came out to say that their technology seems to be three year ahead of all the competitor, certainly including your company. Do you have any comment on your minimal metal pitch and the gate pitch comparison versus Intel? Or do you have any comment for your 5-nanometer versus Intel 10 nanometers potential 7-nanometer?
Well, that's a tough question. I think every company right now, they have their own philosophy developing the next generations of technology. As I reported, in a foundry, we work with our customer to define the specs that can fit their product well. The minimal pitch to define the technology node, we are compatible to the market. The most important is we are offering the best solution to our customer's product roadmap, and that's what we care for. I don't compare that really what is the minimal pitch to define the technology node.
As a follow-up question, sir, can we say your power consumption efficiency in your 5-nanometer in 2020 will be ahead of all their competitors, including Intel, IBM? Can we say that?
We have confidence to do that. Yes.
Thank you.
Next question will be coming from UBS, Bill Lu.
Hi, good afternoon. Thank you very much. Question for Dr. Liu. You said that AI ubiquitous computing will be a driver starting in 2020. I am wondering if you can talk to us about why that timing, because you said that half of your seven nanometer customers are HPC related. It seems like your customer's customers are pretty bullish right now as well. Why that timing? Thank you.
Well, we look at our customer tape out at 7-nanometer, and look at our development activity in our 5-nanometer. I think that is the timeframe that our customer can have a strong advantage to get into that market and grow.
I guess I'm not so clear because 7-nanometer starts ramping 2018, 2019.
Yes. 2019.
Seven plus
seven plus. In order to accumulate huge volume revenue for us, it takes a couple of years cooking to bring the whole value up.
Okay. You're saying it becomes a bigger driver by 2020?
Yes. By driver, I measure by the incremental revenue.
Dollars. Do you have an estimate for how much as a % of revenues or total size by 2020?
Of HPC? I mentioned last conference, it's currently 15%, and at this point, it slightly increases this year by a couple %. Really, it is in the cooking from today to 2020, and we do have a number, but currently is a model. I really cannot be certain on that number's accuracy. The trend is picking up after that time.
Thank you. My second question is related as well, which is, it seems like every quarter you're getting more confident on 7-nanometer, both in terms of your own execution and your customers' progress. I know the comment previously had been that CapEx stays relatively flat next several years. Does that change at all with the confidence that 7-nanometer going out between now and 2020? Thanks.
Yeah. As the program development continues progressing, our confidence gets more, and our customers' confidence also gets more. In terms of the CapEx, of course, we try to make the 7-nanometer and 10-nanometer relatively compatible. We try to maximize our CapEx efficiency when we starting the ramp in the 7-nanometer.
Can I just ask a quick follow-up? If you look at '28, which I think was the last dominant node for TSMC, that coincided with the smartphone ramp between, I'm throwing out numbers, 2011, 2015. Your CapEx went from about TWD 3 billion to about TWD 10 billion, right? That also included a huge revenue ramp. Now, as we look at HPC, I mean, to me, it's a big opportunity as well, and that's over the next several years, but you're not saying that CapEx is going to triple, right? What should we expect?
You're talking about CapEx after 2020?
Bill Lu, you were thinking we should add up CapEx as we be more confident on the technology development. Actually, I want to mention one thing is, actually we have spent tremendous effort trying to improve the capital efficiency. You can see the equipment generation to next generation, the conversion ratio is pretty high. I was talking about 80% that time, people feel it's good. Now I can tell you the 10-nanometer transition to 7-nanometer, the migration compatibility is more than 95%. That's the effort we have spent. We believe the CapEx guidance we are talking about at a $10 billion level has reflected our confidence and all the effort we have put in to improve the efficiency.
Thank you.
Next question will be coming from Citi's Roland Shu.
Hi, good afternoon. First question to Mark. Mark, you mentioned on the Arm-based processor for HPC, for data center HPC actually, demand increased a lot. Do you see the same trend for this Arm-based processor for PC? Yeah, because Microsoft actually-
Yeah. Right
now their Windows 10 is supporting this Arm-based processor.
You're referring to Microsoft Windows 10 supporting the Arm-based-
Arm-based CPU
CPU in the PC and notebook last year, right?
Yeah.
That was very good news to us. I'd rather not to reveal our customers' booking individually. I think we see their business to grow, and they get all the support from us and many other OEMs.
Yeah. We can assume the very strong second half momentum, definitely we probably included some momentum for this Windows on Arm there. Can we assume that?
I didn't say that, but.
Okay.
Yeah. It will grow.
Okay. Thank you.
Timing, I cannot reveal to you.
Okay. Thank you. Also, same question for your 22 ULP. You said you are going to mass production for next year. When can we expect this 22 ULP revenue to achieve 10% of total revenue?
I did not look at that. All I can say is, we start to engage with many customer, and it's very competitive. I expect some of the 28 nanometers product will move into 22. When we are going to reach 10% of wafer revenue, I did not see the exact schedule yet.
Actually, we don't separately looking at 22 and 28, because it's an extension of the technology. We look at a combined revenue. We don't look at it separately.
Okay. Going forward, that means we will include 28, 22 as the same node.
how about 16 and 12? That will be also into the same node.
Okay. Thank you.
next question will be coming from Goldman Sachs, Donald Lu. I will need to translate that first. Donald, so if I misrepresent you have to let me know. Donald's question basically is with respect to the implementation of AI in smartphones, and he's asking whether or not this increment of AI functionality to the smartphones will increase the die size of the smartphone.
It should, every company have their different product strategy. Some target to the low end, some target to the premium. They're all very different. For the low end still, the cost is still critical, and they want you just to get the wows from the customers. they have different strategies that I cannot really quantify for you. Definitely, certain piece of silicon will be dedicated to AI, and in the beginning, could be a separate function and it could be integrated, too. remember, our chip size will shrink along the progress of technology. even maintaining the chip size already is a huge capability build and contribution from AI already.
Maybe I can follow up on TSMC's smartphone content question. For this year, I think last year, last time, six months ago, you commented it will increase for this year. Is that still the trend, and can you repeat how much it will increase from last year to this year again?
Roughly, if you calculate from our share and our average dollar per phone, it's about high single digits.
This trend you think will continue because of AI and other things for next year?
I hope so.
Okay.
I hope it will be more.
Great. My second question is on the guidance for this year. I think in January you said 5%-10% in $, and now the whole market is slowed down a little bit. We will be more in the low end of that range for this year?
No.
Great. Thank you.
Next question will be from J.P. Morgan's Gokul.
Thanks. My first question, just wanted to refer back to your comments on 10-nanometer/7-nanometer being a bigger node than 16. Could we also compare that to 28, which was at, I think at its peak, like $8.5 billion in terms of revenue? What is our degree of confidence in terms of 7-nanometer and 10-nanometer combined at its peak rate being significantly higher than 28? Maybe also you talked about half of the tape outs being HPC related. Could we also have some idea about, is the revenue still going to be mostly mobile related or smartphone related? Is the revenue also going to be more or closer to the split that we see in the tape-out side in terms of HPC?
Well, in Taiwan's right now, to be honest, many of our customer is still working on their products with our 10-nanometer and 7-nanometer. What we do is just to try to bring their product to the market, and how successful that is still really is yet to be seen. It's very difficult for us to calculate what the business, and how many wafers or how many product units their product gets to achieve. I can tell you that in dollar sign, to us, it's a much bigger volume, much bigger in nodes than either 16 and 28. Okay. For the same reason, it's how much is the HPC, how much is the mobile computing. I think on 10-nanometer or 7-nanometer, it's still too early to say. HPC will be the bigger portion than you have been known. Yeah.
Okay. My follow-up is on N7 and N7+. Since there is additional EUV steps, do customers have to redesign their product? Does it drive more customers towards N7+ rather than N7, or you're not seeing any of that happen?
They, too, have to recharacterize the standard cell, something like that, but the effort will be very minimum because we try all our best and work with IP infrastructure partner to minimize that effect. It will be, I cannot say transparent, but the effort will be minimized.
You don't see any customers delaying their N7 plans because they want to wait for N7+ with EUV? You don't see that?
Well, I think the customers has their own product roadmap, their new product introduction to the market is probably the most important thing for them to consider. Whether they're delaying their product from N7 to N7+, we did not see that yet.
Thank you.
All right. We will now take our next question from the call. Operator, please proceed with the first caller on the line.
The first question from the line comes from Brett Simpson from Arete Research.
Yeah. Thanks very much. I have two questions. First of all, on the HPC segment, if I look at that segment and look at chips like data center chips, server chips, this HPC segment is structurally much higher margin, and much more profit in the value chain. Do you think TSMC can make structurally higher gross margins in HPC? Maybe you can discuss that opportunity. Just on the earlier questions of evolving customers to 22-nanometer and 12-nanometer ULP, you mentioned there's minimal upfront NRE for customers to evolve their chips from 28 and 16. Can you talk a bit more about this? It seems to be quite a big differentiator versus a customer going to FDSOI or another fab. Can you maybe talk about how significant a saving this is for customers? Thank you.
Okay. Let me answer your HPC's, how to capture values. Yes, you're right. The HPC products produce much more value in the system build. Our customer will command more value products. We work with our customer, also will trying to get more value along the supply chain. Remember, this HPC product will not only be wafer business. All our advanced packaging technology and associate with our future platform of system integration will also come to the service provide to our customers. In the future, the advanced packaging, we also intends to increase our value to our customer, therefore capture more value along the supply chain.
Well, let me answer the question on 22 ULP and 12 ULP, how competitiveness versus our competitors are offering the 22 FDSOI or others. The first one on the 22 ULP. This one we offer lower power consumption, higher speed, and with the die size shrink about 10%, as I said. The beauty is our customers don't have to change anything. They can use their IP portfolio in 28-nanometer. The 10% die shrink is actually is a optical direct shrink. The customer does not have to do anything. They can easily port their existing product or the new product into 22 ULP and take advantage of the higher performance, and most important, much lower power consumption. Then for their new product to compete in the market.
We think it is very competitive. On the 12 ULP, we can shrink a little bit on the metal pitch, but as I just mentioned, we also work with the IP infrastructure partner to minimize their effort. Again, this one offers a very effective path and cost-effective path from 16 nanometer into 12 nanometer. We expect for this kind of advantage we offer to our customer, we still can maintain a very high market share.
All right. We will still stay on the line and take the question from the next caller. Operator, please.
Next question is from Steven Pelayo from HSBC.
Thank you. A year ago, we talked about your lead customers at 16 nanometer being on 16 nanometer for the second year. I believe it was CC that said, yes, they're increasing the functionality there, so the die size doesn't really change too much there. This year, as we go to 10 nanometer, we get a little bit more benefit from a stronger shrink. I'm curious, what do you think are the impacts? Will a smaller die size thus require less wafers from you in the second half of the year? Or more importantly, the ASP differential at 10 nanometer significantly more versus, let's say, 16 nanometer FFC, that it more than offsets any potential die shrink. Could you comment a little bit on die sizes and maybe ASP differentials on 16 nanometer FFC versus 10 nanometer?
All right. I will repeat Steven's question. He is basically asking that our major customer this year will migrate from 16 nanometer to 10-nanometer, therefore, he assumed that it will be a smaller die. Since it is a smaller die, whether or not it will be fewer wafers. If it is fewer wafers, whether or not the ASP increase can more than offset by the fewer wafers.
Well, we don't comment on customers' die size, first. Second, I think Mark already pointed out, as we move along the path of the technology, the geometry gets smaller and smaller. However, the content also increases quite a lot. You can see that from your smartphone you are using today and two years or five years ago. Again, I don't comment on the customer's die size.
Steven, do you have second question?
I do. Well, the second part of that first question was relative to the ASP differentials on 16 nanometer FFC versus 10-nanometer.
Oh. He wants to know how much more we charge at 10-nanometer compared to 16FFC.
Confidential information.
Okay, operator, let's move on to the next caller on the line. Thank you.
Thank you. Next question is from Patrick Liao of Macquarie. Please go ahead.
I'd like to ask your view of entire year seasonality. Is it changing gradually from recent years, and the first half has become relatively sluggish? This is my first question.
Patrick's question is, what has changed in terms of our business seasonality? It looks like our first half is sluggish compared to second half.
Your observation might be right. This is the second year that we have a slower first half, and that we, at this point, look at it has to do with the premium phones seasonality, as I have mentioned. We are going to develop more customer products, hopefully, to smooth out this. This is the current seasonality we see.
Okay. My next question would be, TSMC moves 3-nanometer fab plan to U.S. I know this probably not right to ask, but very curious, I assume, for many audience. Thank you.
Patrick, you are asking if we will go to U.S. to build a 3-nanometer fab? Is that your question?
Yes.
Okay.
Yes.
We continue to look for the options on the new fabs location. So far, the fab in the U.S. is not very optimal due to many considerations, although it's still an open option for us to choose.
Okay.
Okay.
There seems to be quite a few people still on the line, so we will continue on the line. Operator, please have the next caller. Thank you.
Next question is from Mehdi Hosseini from SIG. Please go ahead.
Yes. Thanks so much for taking my question. Two follow-ups. One on EUV. Can you please help us understand the extent of EUV adoption? Specifically, how many layers would you expect the N7+ node to adopt EUV? I have a follow-up.
Well, I say a few critical layers, which we use to do two purposes. One is to swing the die and then to do the practice for the EUV. How many critical layers? That I cannot share with you now.
Did you say a few critical layers?
A few critical layers.
Thank you. My follow-up question has to do with the capital intensity and how you have been able to increase efficiency of equipment. I want to better understand the equipment reuse from 10-nanometer to N7. I heard Lora talking about a TWD 10 billion CapEx for a few years. I also heard the equipment that is used for N7+ is going to be pretty much the same set or very much reusable from 10-nanometer. Can you help me reconcile the CapEx with the equipment reuse commentary?
You're asking capital intensity and how do we reuse the tools from generation to generation. I have said maybe a couple of times, in the coming few years, we expect our capital intensity will be ranging from 30%-35%. I'm still with that view. In terms of how do we do it, because capital is very expensive in the company, and that's the utmost important decision in the company. We spend a lot of effort trying to As the generation becomes shorter, so that make the equipment reuse even more critical. We have invented a lot of way, trying to increase the commonality from node to node. Either if some idle or open tool, we're trying to utilize those idle open tool. Either migration or mutual, several fab back up each others.
A lot of innovation coming from the operation people, trying to reduce the CapEx we have to spend. Also, we are putting more focus on what's the peak capacity building. You cannot just follow what customer tell you, the highest demand they need. You have to look at the cycle, and find out which the optimal peak capacity you build. Once you build it, you try to utilize the capacity as much as possible. Including, sometime you have to move product among quarters, just for purpose to utilize the capacity. That are the several ways that we have been doing the company to improve the capital efficiency.
Can I ask a quick follow-up on that?
Okay.
Just very quickly, as you scale the InFO backend technology, should we expect the mix of CapEx between front-end and backend packaging change going forward?
Mehdi's question is, now since we have started volume producing with InFO, what will be the mix going forward between the capital spend for the front end, which is for the wafer, and then the capital spend for the back end, which is the packaging.
Mehdi, it is indeed, our backend capital has been increasing. For example, we are ramping the InFO technology starting from last year, and I didn't show you the CapEx breakdown. It's pretty high. In the past, I think our backend is a TWD 200 million. It can go up to TWD 1 billion, and this year will be the case.
Okay, I think we'll come back to the floor now. Now we are back to the floor for questions. First, let's have Morgan Stanley's Charlie Chan.
Thanks. My first question is on the second quarter guidance. Mark, you mentioned that there were some supply chain inventory management. Besides the smartphone, what area do you see a weakness? For example, gaming, PC, even industrial. Would you also attribute the smartphone chips weakness to your customers' market share loss, meaning loss to Qualcomm? Thanks.
The second quarter weakness, smartphone is one. We think the PC is another one. Mostly because of inventory though. Industrial, I don't see the weakness at all. Come to your question about the What's your second question?
Yeah, your customers' market share loss.
Oh, okay.
to its competitor.
Okay. This is not news, right? We anticipate this, and we are working to recover for quite some time. All these years, recovery work is ongoing. We already factor those in. Still, we expect this year, overall, we should be able to gain market share overall to overcome this. For the long term, of course, the relationship with this customer gets much better, warmer now, and we are working with them on the future technologies.
Okay, thanks. Also, on your HPC- ready COM, you mentioned that you are excited about Arm-based server get qualified. Does that mean you already have that second win for your 7-nanometer? The Arm-based server processor in your 7-nanometer customer already? Yes.
I'm sorry, the question is?
Do you already have Arm-based server processor-
Oh, yes. Of course
already?
Of course. Yes.
Okay, thanks. My next question is to Mr. Wei. Regarding your 28-nanometer capacity expansion, I think a couple of weeks ago, at your technology symposium, you mentioned that 28-nanometer capacity will increase 15% year-on-year. I don't doubt your competitive of 22ULP, but the fact is that some of your customers are moving to 16 nanometer, and also your market share at 28 nanometer is already more than 90%. How confident that your demand can grow another 15% in the coming year?
Well, we build our capacity according to customers' demand, and we do it very carefully. You say that we are going to increase 15% every year? Did we say that? This year. Oh, this year. Oh, this year, yes. So far, I can share with you that we are still fully loaded and probably a little bit not enough. We work very hard to increase the capacity to fulfill customers' demand.
Even with China's smartphone weakness, you still think your 28 nanometer is fully loaded?
Yes. It's continued to be that.
Okay. Yeah, very quick on just one number. You mentioned that your 7-nanometer customer already exceeded 30 customers. What's the customer number for 10-nanometer and 16-nanometer respectively currently? Thank you.
Well, probably comparable to 16-nanometer, but the 10-nanometer is smaller.
Okay.
The 7-nanometer and 16 are similar.
Okay. Understood. Thanks.
Okay. Next question will be from Daiwa, Rick Hsu.
Yeah. Hi. Good afternoon, and thank you so much for taking my question. Just a quick follow-up to the previous question about your second quarter demand driver. I think Mark mentioned about the weakness mainly from smartphone and PCs due to some more inventory question, and industrial tend to be stable. Do you see any pocket of strength for second quarter?
I don't particularly notice strengths in certain pockets. Industrial is considered heating up. That's the number showing the overall second quarter. Okay. Automotive, although a small percentage, is still heating up.
Okay. Thank you. My second question is, if I remember correctly, I think your 28 nanometer is a very strong node, and even for this year, you still constantly build up more capacity. For your 20, 15, and 10, I think capacity-wise, for each node is much lower than 28 nanometer. My question is, do you anticipate that next seven and seven plus will be a strong node with capacity build, maybe by 2020, it will be at least as comparable as your 28 nanometer?
I cannot share with you, but I certainly hope so.
Okay. Thank you so much.
Okay. Next question will be coming from Credit Lyonnais, Sebastian Hou.
Thank you, Dr. Sun. I think my first question is to Dr. Liu on the HPC. Can I follow on that you mentioned by 2020, HPC will become the bigger driver for the company? Is it bigger or biggest? In terms of incremental revenue, right?
Well, that depends on how smartphone will play out then. It's a competition. I think in 2020 to 2025, that's the time where they really picked up. I wouldn't say in 2020 to compare with the smartphone yet.
Okay. Just to follow up on that one is that I think 6 months ago, that conference, you gave us some idea about the wafer addressable market revenue. HPC is about $15 billion US by 2020. I remember at that time you mentioned about it's really hard for you to quantify artificial intelligence in the numbers. I wonder, the thing that you talk about today, is it on top of that $15 billion or already included?
It's part of the story. It's still a consistent story. When I talk about AI, I talk about HPC, but I also talk about IoT, automotive, and smartphone. AI is a phenomenon of using more silicon in all kinds of devices. Of course, the AI will require more computation in the data center and more communication in the network and storage. That HPC is part of this AI portion in HPC application.
Okay. Thank you. One follow-up. I think this one for Dr. Wei. We understand that your 16-nanometer, one of the major customers, moving to 10. With your launch of the 12-nanometer, do you expect your 16-nanometer family with 12-nanometer next year, the total revenue in absolute level will recover to the same level as we've seen in 2016?
We hope so, but let me stress again that 12-nanometer will be used for the need to go in smartphones. We offer a way for our customer to have a cost-effective roadmap for their product. Again, the most advanced and premium grade smartphone will move to the next node as you expect it.
Thank you.
Okay. Follow-up question from Randy Abrams, Credit Suisse.
Yes. Thank you. This one probably for C.C. On the Fan-out, you talked about more mobile products. Could you talk if that's still for the high-end smartphone or you see it moving down into the mass market where it's now cost competitive with Flip chip? Do you see any cases where customers are splitting the architecture now? Using InFO and maybe migrating a smaller part of the die size, say the seven and 10-nanometer, and then with InFO keeping some on lagging nodes.
Okay. Far today, InFO are still used by most advanced smartphones. We are engaging with the customer, many, to do the things you just mentioned to move into a more area, including separate die size into a multiple one to get more efficiency and roll down the cost. That's what we are doing. More importantly, we are moving the application into the High-Performance Computing because Mark just mentioned the AI and all those kind of thing will start to develop in the next few years, and we try to catch up with the market.
Okay, great. The second question I wanted to ask. You maintained, aside from currency, your growth assumptions for this year, but you lowered the overall foundry industry. Could you maybe talk about factor? Was it a difference in share? You feel like you've taken some share in certain applications or a view on what you expect your competitors versus a few months ago?
Well, the main reason we revised our foundry growth rate is due to the inventory accumulation we see in end of Q1 and also expected towards the end of Q2. Our forecast currently, however, is better than the average foundry growth rate forecast. Therefore, we consider we are gaining market share this year.
Just to clarify, from this inventory correction, you think you've gained enough share to offset, say, for TSMC, the impact of this inventory correction?
Well, inventory correction applies to everyone.
Okay.
Simply that therefore, the average foundry market share will drop. Will we, if we are not gaining other share to be better than 5%.
Thank you.
Next question will also be coming from a follow-up, and that will be Citi's Roland Shu.
Thank you. I just look at your fourth quarter report, and you have this annual result. Last year, your non-wafer revenue actually declined in both absolute number and also as a percentage of the total revenue. Even I look at in the past year from 2013 to 2016, the percentage of the non-wafer revenue actually declined from 6.1% to 4.1% last year. I think this actually is, again, because in the past years, actually, we had a lot of activity, to doing the 20 nanometer, 20 or 16, 10 nanometer, means a lot of the photomask or revenue layer. Can you explain what's the disconnection with a lot of the new wafer activity, but we still have this declining non-wafer revenue in the past years. Thank you.
I probably cannot link your statement about the declining revenue. There is a period of time our non-wafer revenue, basically the back-end and the EBO, that's the two major non-wafer revenue. It has been about 10% for many years. With InFO that we're joining the company starting from this year, we expect at least we can maintain or even a small increase on non-wafer revenue, as far as I can see.
Yeah. The 10%, I think that that is probably too big a number, compared to what the number you reported. That in last year was only 4.1% of the non-wafer revenue.
Cannot be. I have to check which page you're looking at.
Page 80 something.
80 something. I'll go back and check.
Okay, you have the wafer revenue and the non-wafer revenue there. Yeah. Also a second question is for the InFO. I think, Dr. Wei guided us for InFO revenue last year, 4Q last year was above $100 million. How about, your forecast for InFO revenue this year? Thank you.
I mentioned, this year you will be, for the 2017, you are about $500 million.
All right. Follow-up question from UBS, Bill Lu.
Thank you. Two quick follow-ups. One is on Donald's question on smartphone content. In the past, you've given us the content for high-end, mid-end, and low-end. Can you provide that for this year?
I don't have that with me. The similar situation than last year, I think.
Okay, great. Thanks. Just a clarification with Dr. Wei. You had said that HPC will increasingly use InFO. I had the thought that maybe a lot of the HPC applications will use CoWoS. Can you talk about the two?
Oh, yeah. Today, almost all of the HPC customer using the CoWoS because of that one is a very high performance. I'm talking about extend the InFO into the HPC area because that offer the lower cost. Also, we improve the InFO's performance. The performance still lower than the CoWoS, but cost effectiveness is much better. That's what we expect to be widely used by HPC customer.
Okay. Follow-up questions from Deutsche Bank, Michael Chou.
You mentioned the HPC for InFO in the future. For very high pin count HPC product, would that still use CoWoS, or you think they will shift to InFO entirely in the future?
No, I would think it is still using CoWoS. That is still the much better performance for very high pin counts in the few thousand contact or those kind of things.
Can we say it maybe 70%-80% HPC product will be based on InFO, the new InFO product in the future? Or you think that maybe 50% of the HPC will be still based on CoWoS?
No. Today, I cannot do any prediction. We are working with the customer because of overall cost structure is very important. That's what we are working on. We also try to make a lot of variation to meet the customer's requirement.
Yeah. As a follow-up for your InFO, do you think the mid to low-end product will shift to InFO because the things that mid to low-end product care about cost?
I would expect that. Yes. It's not happening yet.
In the future, it is possible?
It's possible.
Okay. The second question for Mark. The management mentioned before, you should have a higher market share at 7-nanometer versus 16/14, right?
Yes.
Can we say that you can gain more market share in 7-nanometer smartphone market versus 16/40-nanometer in 2018-2019? I'm not mentioning the customers. I'm just mentioning market share.
Yeah. I think so. We will have higher market share in 7 and 7+. Those are for high-end smartphone. The mid low-end smartphone, we still working on 22 ULP, all those low power technology to capture. We have a multiple-prong strategy on this to capture share.
Okay. Questions will be coming from JP Morgan's Gokul.
Thank you. First question, when I look at HPC currently, unlike smartphone, like you mentioned, premium will be at seven nano and 10. Older nodes can also be driven by smartphones because of low cost or mid-end. HPC, if I use the current biggest IDM, which is probably the biggest HPC player right now, most of their capacity is leading edge, and they probably have N and N minus one, that is pretty much it, and they work on a very accelerated schedule in terms of capacity conversion.
When HPC becomes a bigger driver for TSMC, how do you think about capacity planning? Are we going to have a lot more conversion from, let us say, 10/7 to 5? Or are there enough second wave, third wave, even within that ambit of ubiquitous computing that could drive second or third wave for the fresh capacity that you put on?
You are talking about sometimes in the future. Let me ponder this with you. TSMC is very different than this IDM company. We are in every segment in the market. Therefore, we attack the leading edge, and we capture the HPC. Does not mean those technologies later on, when the technology moves, it does not have the usage for all other segments. Let alone you talk about HPC's volume compared with today's premium phones, is still smaller.
Okay.
We will consider the technology migration, capacity migration then, but I don't consider that a major change or major problem for us.
Okay. My follow-up is on a more near-term question, especially for second half. If I look at the half-on-half growth, it's almost 20% going into the second half. I just wanted to understand, are we anticipating that except for the big premium customer ramp-up, are we also anticipating some inventory restocking? Are we expecting inventory correction to finish in first half, and we anticipate some general inventory restocking in second half? Or is it primarily driven by the super cycle that people are expecting?
Yes, we try to prepare for the flexibility to cope with the demand of the second half. You're right, certain activity is already ongoing in the second quarter to make sure that we can deal with the needed demand in the second half.
If I could phrase it a little bit differently, do we expect the customers who are really weak in the mid-end smartphone, et cetera, also to have a big rebound in the second half of the year, along with the big seasonality for your main customer?
The smartphone high mid, low-end, we consider is we can deal with it because at this point, it still follows this relatively seasonal patterns, and it is more predictable than the premium launch.
Okay. Follow-up question from Goldman Sachs, Donald Lu.
May I ask a question on other gains and losses? It is still fluctuating. Is there a guidance and where it is coming from, et cetera? It is TWD 2.62 billion for Q1.
Which line you are asking?
Other gain and loss. I was looking at page three of the management report. On the second table there is other gains and losses was like TWD 1.8 billion in last Q1 and TWD 2.6 billion this Q1. I was just wondering.
You refer to 2.62?
Yeah.
Second quarter versus the TWD 1.8 billion last year, right?
Yeah.
I think a big component of this line is the interest income. If we have more cash, it generate more interest, and that help on this one.
Okay. Another question is on this year's guidance again. I think you mentioned the lower foundry forecast because of inventory correction. Have you considered the news or speculations now that one of the high-end premium phone launch this year might be delayed? For example, could be launched in September, then shipment is in November, December, something like that. If that happens, what will happen to TSMC's guidance?
This is totally speculation, right? I don't make comment.
Okay. In the interest of time, we'll just allow the final follow-up, and that will be from CLSA, Sebastian Hou.
Thank you, Dr. Sun , for choosing me. The first follow-up I have is, we've seen TSMC blended wafer price on the 12-inch equivalent basis has increased every year since 2010. My question is that, are you confident in terms of continuing to increase your blended wafer price in the next five years?
So far we're doing okay. We're still trying hard to continue improve our ASP. That's all I can say. Yeah.
Thank you. My second follow-up is, can you share with us your strategy and how you look at memory? We understand that TSMC has been developing embedded memory, also that a new type of the innovative memory like MRAM, RRAM. Can you share with us how you progress on that, and when we're going to see that embedded memory technology being commercialized on 16 or even 7 nanometers? Also, can you share with your strategy on the discrete memory?
Let me talk about discrete memory, and C.C. will cover the embedded memory. We consider we are not in commodity business. Therefore, at this point, we do not intend to enter the commodity business currently, which is discrete memory is.
Other than the commodity, actually, we work with customer to offer them the embedded memory because of, today, we are the largest provider of embedded flash. embedded flash has its own limitation after 28 nanometer. We work with the customer to find a better solution. That means is we have to find some memory that can dissipate low power and higher speed. That's why we are working on the RRAM, MRAM, and others. Okay. That's our strategy. We certainly work with customer to meet their demand, and we define our embedded memories roadmap.
All right. With this answer, we conclude today's conference. Please be advised that the replay of the conference will be accessible within three hours. Transcript will become available within 24 hours from now. Both of them will be available through our website. Thank you for joining us today. We hope you will join us again next quarter. Goodbye and have a good day.