[Foreign language] Welcome to TSMC's second quarter 2016 earnings conference and conference call. This is Elizabeth Sun, TSMC's Director of Corporate Communications, and your host for today. Today's event is webcast live through tsmc.com. If you are joining us via conference call, your dialing lines are in listen only mode.
As this conference is being viewed by investors around the world, we will conduct this event in English only. The format for today's event will be as follows. First, TSMC Senior Vice President and CFO, Ms. Lora Ho, will summarize our operations in the second quarter of 2016, followed by our guidance for the third quarter of 2016. Afterwards, Ms. Ho and TSMC's two Presidents and Co-CEOs, Dr. Mark Liu and Dr. C.C. Wei, will jointly provide our management's key messages. We will open both the floor and the line for the Q&A. For those participants on the call, if you do not yet have a copy of the press release, you may download it from 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 statement that are subject to significant risks and uncertainties, which could cause actual results to differ materially from those contained in the forward-looking statement. Please refer to the safe harbor notice that appears on our press release. Now I would like to turn the podium to TSMC CFO, Ms. Lora Ho, for the summary of operations and current quarter guidance.
Thank you, Elizabeth. Good afternoon, everyone. Welcome to joining us today. My presentation will start with financial highlights for the second quarter, followed by the guidance for the third quarter. We had a good second quarter. Our second quarter revenue increased 9% sequentially to TWD 222 billion, exceeding the high end of our guidance given in April, due to business upside resulting from the demand increases in mid and low-end smartphones and the customer inventory restocking. Gross margin was also ahead of expectation and increased by 6.6 percentage point quarter-over-quarter to a record high of 51.5%, mainly driven by improvement in utilization rate and cost reduction efforts, partially offset by an unfavorable foreign exchange rate. Operating margin also increased by a similar magnitude to reach 41.2%. As we highlighted last quarter, we incurred a higher income tax in the second quarter.
The corporate tax rate rose to 23% as we accrue the 10% on undistributed retained earnings. We estimate the 2016 full year tax rate to be about 14%. Overall, our second quarter EPS reached TWD 2.80, an 8.7% decline year-over-year, mainly due to a non-recurring share disposal gain from ASML and Vanguard in the second quarter of last year. Excluding the one-off items, second quarter EPS actually grew 17% year-over-year. Let's take a look at revenue contribution by application. During the second quarter, communication was flat from the prior quarter, while computer, consumer, and industrial and standard increased by 19%, 18%, and 12% respectively. How about revenue by technology? Combined revenue from 16 nanometer and 20 nanometer continued to grow and represented 23% of revenue in the second quarter, same as the first quarter.
28 nanometer contribution fell slightly to 28% versus 30% in the first quarter due to a higher revenue base in the second quarter, but absolute dollar revenue from 28 nanometer continued to increase Q-over-Q. Moving on to the balance sheet. We ended the second quarter with cash and marketable securities of NTD 668 billion, an increase of NTD 19 billion. On the liability side Current liability increased to NTD 163 billion as we accrued about NTD 156 billion for cash dividend, which will be paid out this month. On financial ratios, accounts receivable turnover days increased two days to 43 days, while days of inventory remain at 54 days, same as last quarter. Let me make a few comments on cash flow and CapEx. During the second quarter, we generated about NTD 107 billion cash from operations and spent NTD 74 billion in capital expenditure.
As a result, we generated free cash flow of NTD 33 billion this quarter and repaid NTD 11 billion of corporate bonds. Our overall cash balance increased NTD 4 billion to NTD 622 billion at the end of the quarter. In US dollar terms, the capital expenditure spent in the first half of 2016 totaled $3.4 billion. The second half spending will be much more than the first half. I have finished my financial summary. Let's turn to the third quarter outlook. We expect our business in the third quarter will benefit from a customer's new product launch and continue inventory restocking. Based on our current business outlook and exchange rate assumptions of one US dollar to NTD 32.30, we expect third quarter revenue to be between NTD 254 billion and NTD 257 billion, which represents 14.5%-15.9% sequential growth.
Gross profit margin to be between 50%-52%, and operating margin to be between 39.5%-41.5%. This concludes my remarks.
We will begin the management comments, and we will start with TSMC CFO, Ms. Lora Ho.
I would like to make a few comments on near-term outlook, CapEx, profitability, and the dividend. Let me start with near-term outlook. Our second quarter result was helped by an increase in demand from China 4G plus smartphone ramping and continued 3G to 4G upgrade from emerging markets. Given a stronger than seasonal business for our fabless customers in the second quarter, we estimate our fabless customers' DOI exceeding second quarter is above seasonal level. Looking ahead to third quarter 2016, we forecast continuing inventory build by the smartphone-centric OEMs and fabless companies stimulated by subsidy provided to the telecommunication companies by the Chinese government. We estimate fabless DOI will remain flat with the second quarter, but will be above seasonal level by a few days, exceeding third quarter. Fourth quarter is normally an inventory adjustment quarter, but the degree of adjustment will depending on market dynamics and sentiment.
We expect the supply chain inventory to go back to the seasonal level exceeding fourth quarter. For the whole year of 2016, TSMC maintains our target of between 5% and 10% growth in both revenue and operating profit, both in U.S. dollar and in TWD term. Let me make a few comments on CapEx. Our 2016 CapEx forecast has risen from $9 billion-$10 billion to $9.5 billion-$10.5 billion because expectation for 2017 mobile products revenue has risen as well. TSMC's CapEx to sales ratio, known as capital intensity, has come down significantly in the last two years. Compared with the high 40s level seen in 2011-2013, our capital intensity has dropped to about 31% last year. Going forward, we estimate our capital intensity will remain at mid-30s level for the next few years.
One major factor contributing to this moderate level of capital intensity is our effort made to minimize the conversion loss between two adjacent technology node. For example, between 20 nanometer and 16 nanometer, we closely manage one peak capacity so we can minimize conversion loss. We also closely manage 10 nanometer and seven nanometer as one peak capacity to minimize the conversion loss. In terms of profitability, in the past few years, despite the higher CapEx, which led to a substantial increase in depreciation expenses, we have been able to maintain and even improve our structural profitability. We expect to be able to maintain our growth margin at close to 50% level and operating margin at close to 39% level. My last comment is about dividend. We anticipate a steadily improving free cash flow in the next few years.
Therefore, we expect our dividend to steadily increase in the next few years as well. Thank you. Let me turn the microphone to Mark for his comment.
Good afternoon, everyone. Thank you for joining us. I'd like to deliver you the message firstly on leading edge technology status that include 10nm, 7nm, and 5nm. First, on 10nm. Our 10nm has been transferred from R&D to production. Our first 10nm customer product has been produced with satisfactory functional yield. So far, three customer products have been taped out to us. More customer product tape-outs are expected later this year. Those product tape-outs will start a revenue stream starting first quarter 2017, which will ramp steeply throughout 2017. On 7nm. Our 7nm technology development is well on track. Its 256 megabit SRAM yield improvement is ahead of our schedule. In addition, we believe our 7nm PPA, that is power, performance, and area density with its schedule is ahead of our competitors.
This technology has been aggressively adopted not only by mobile customers but also by high-performance computing customers. They all have aggressive product tape-out planned in first half 2017, with volume production planned in early 2018. On 5nm, we have been executing our 5nm development since the beginning of this year. TSMC's 5nm will achieve 1.9 times of logic density over our 7nm. We plan to extensively use EUV lithography in 5nm to improve density, simplify process complexity, and reduce cost. The 5nm risk production qualification in first half 2019 remains unchanged. Secondly, on EUV status. We plan to adopt EUV extensively in our 5nm technology. Today, EUV technology uses 7nm as a development vehicle. We have good integration progress in EUV scanner, EUV mask, and EUV photoresist. Same CDU level has been repeatedly demonstrated using two EUV layers in 7nm.
Currently, we are running four state-of-the-art EUV scanner for EUV infrastructure development and for N7 and N5 technology development. We will move in another two EUV high volume production tools, that is NXE 3400, in first quarter 2017 next year. Recently, we successfully implemented 125 watt EUV source in our EUV 3350 to improve productivity. In our in-house mask shop, we have developing unique EUV mask technologies on mask blank, material, inspection, and repair to seamlessly integrate our EUV lithography total solution. With all the encouraging development and progresses made at EUV, we estimate that EUV will be cost-effective tool for high volume manufacturing by 2020, in time for our 5nm ramp. Should EUV become cost-effective earlier, say around 2019, we believe we can still benefit from the earlier available of EUV for our 7nm high volume manufacturing. Thirdly, on growth drivers. First, smartphone as a growth driver.
We expect smartphone-related demand will support half of our growth in the next five years. It will come from unit growth, as well as increasing silicon content. Despite a weak semiconductor industry growth this year, we estimate the smartphone unit growth rate will be about 6%. The silicon content growth rate per smartphone will increase by a double-digit for high and mid-end, and about flattish for the low end. We estimate total semiconductor revenue, excluding memory, in smartphone, will increase about 7% this year. Longer term, we believe the smartphone unit will continue to grow at a mid-single-digit rate, and the silicon content will continue to increase. This silicon content increase is driven by the increasing adoption of innovative smartphone features, such as dual camera, security sensing, augmented reality, virtual reality, and migration of 4G+, and to 5G.
Most of the high-end smartphone features are also proliferating into lower-end smartphones. Because those innovative features usually require more advanced technologies. With our customers, we will gradually increase our market share in the smartphone market. As to high-performance computing as a growth driver, as the digital trend develops in all industry worldwide today, silicon IC will be needed to take up vast amount of computation load in the future. We estimate computing opportunities will support about one-quarter of our growth in the next five years. In TSMC, we work with IC innovators around the world. We see momentum building up in the computing space for our customers. In May this year, seven companies announced that they are forming an interconnect consortium called CCIX to enable heterogeneous computing in data centers. In China, cloud leaders and a technology company formed Green Computing Consortium, aiming to develop energy-efficient data centers.
We also see increasing activities around machine learning, where system companies differentiate down to the silicon level. TSMC, as a foundry player, focuses on enabling our customers' innovation through providing leading-edge technology. Other innovations in computing area are augmented reality, virtual reality, gaming, and ADAS, or Advanced Driver-Assistance Systems. TSMC has been developing suitable process technologies and design enablements for all those innovations. Above is my message. Thank you. I turn the microphone to C.C.
Thank you, Mark. Good afternoon, ladies and gentlemen. Let me start with our 16 nanometer status. We continue to ramp 16 nanometer with defect density and cycle time better than our plan. Major applications this year include mobile processor, cellular baseband, graphic, and video game. We are happy to report that we have been recognized by a major customer for our contribution to the success of a deep learning chip using TSMC's 16FF+. We expect our 16 nanometer business will continue to increase in the second half of this year, with most of the products adopting our 16FFC, which is a low-power and low-cost version of the 16 nanometer process. We expect to generate more than 20% wafer revenue from 16 nanometer in this year. 28 nanometer.
Our 28nm has entered its sixth year of volume production. TSMC's 28 HPC and 28HPC+ have been widely adopted by most of the mobile application processor suppliers for the faster growing mid to low-end smartphones and other applications such as Wi-Fi, digital TV, set-top box, flash controller, et cetera. With our differentiated technology, stable yield, short cycle time, and large capacity support, we expect strong demand for our 28nm will last through this year. Now let me touch a little bit on our competitiveness for those technologies which has been in volume production, namely 16nm, 28nm, and so on. TSMC continues to improve our technologies and develop new variance even after this technology had entered mass production stage.
For example, to meet the changing market requirement, we introduce a low cost, low power version 16FFC this year after 16nm started volume production last year. We also lowered the operating voltage of our 16nm process to 0.5 volt to meet the requirement for IoT applications. Similarly, at 28nm node, we introduce 28 HPC and 28HPC+ for the low power solutions while keeping the speed at almost the same level. For 40nm, TSMC also introduce 40 low power for low power application. For 55 nm and older node, we apply similar approaches as well. Since we are able to start volume production ahead of our competition, TSMC has enjoyed a substantial advantage in learning curve. We are able to accumulate enough experience faster than our foundry peers, therefore, we can better improve our cost and device performance for each technology node.
Because of the learning curve advantage, we believe we are highly competitive in both performance and cost. Now let me talk about our another growth driver, which is IoT in addition to mobile and high-speed computation market, which Mark just explained. Driven by innovative business model, we expect the diversified and fragmented IoT applications will become one of the most important growth driver for future semiconductor business. IoT applications require not only the collection of huge amount of data, but also the ability to analyze the data and utilize the data to improve our daily lives, such as security and health. In the near term, we can see smart band and smartwatch for healthcare, and smart meter for efficient energy and cost saving. Longer term, we expect ADAS and autonomous driving be widely adopted. Smart home, smart city, robot, drone are also progressing along.
We believe TSMC's specialty technology development in CMOS image sensor, MEMS, RF, power management IC, and emerging memory are well positioned to capture the IoT opportunities. Now let me talk about our advanced back-end technology. First, on InFO. We are in volume production now. Our focus today is to continue yield improvement and cost reduction. We expect InFO will contribute more than $100 million revenue in 4Q this year with moderate gross margin. At the same time, we are also developing the next generation InFO technology for products planned for next year and beyond. Now on CoWoS, we have seen strong momentum from high memory bandwidth and high-performance driven applications, such as networking, deep learning, and artificial intelligence. While our CoWoS is still in small volume production for those very high speed applications, we see the potential of demand increase as well.
We expect to support our customer to grow their business in the high-speed computing space with our CoWoS as part of our advanced technology offer. Thank you for your attention.
This concludes our prepared statements. Before we begin the Q&A session, I would like to remind everybody 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 from the call. Should you wish to raise your questions in Chinese, I will translate it 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 queue, please press the pound or the hash key.
Star and 1 if you want to ask questions, and hash key if you want to remove yourself. Now let's begin the Q&A session. I think we will have the pessimistic analyst from J.P. Morgan to ask the first question. Gokul.
Thank you very much. First of all, on smartphones, and growth coming from smartphones being half of the growth, and compute also contributing to quarter of the growth. Could you talk about what is the growth picture for the whole company? Are we going to maintain the 5%-10% growth that we are predicting for this year into the next five years? Or, how should we think about the growth pattern? And I have a follow-up on the compute side as well.
Okay. Thank you. You talk about the growth model. Right. Okay. We do have a growth forecast within the company. Let me share you with the big picture. In our model, semiconductor has been stay at 0% growth for the past 2 years. For the next 5 years, we estimate still about between 2% or 2%+ to 3%. The fabless company, because of the business model advantages and the leading-edge technology advantages, they will grow about 5% per year in the next 5 years. TSMC with not only the smart fabless company, but also the system company and IDM outsourcing. Our current estimate will be about between 5%-10%, depending on the macroeconomics and the semiconductor events. This model is assuming that there's no killer application happens in the next 5 years, okay?
As to the background of this growth model, we are focusing on 4 areas as I presented to you earlier. First is the smartphone, mobile devices, mostly smartphones. That will drive about 50% of that growth in the next 5 years. The high-performance computing, that will include the learning, artificial intelligence, the machine learning, AI, gaming. That will drive about a quarter of our driving growth in the next 5 years. Autonomous car and IoT and the rest will take up another quarter of the drive growth for the next 5 years. This is largely our growth model. Yeah.
Just on the smartphone bit. I think we've seen smartphone estimates coming down for the last 6 quarters. Are we expecting that smartphone units stay at this 5%-6% kind of growth rate? Could, I think in most tech markets, once the saturation hits, we get down to probably GDP growth rate, which is like 2% or something like that. Are we anticipating that? Is the future growth in smartphones primarily coming from content growth, which is still pretty decent?
Indeed, the smartphone has been slowing down in the past 6 quarters, particularly for the high-end. At the same time, the mid-end silicon content is increasing very fast. The unit number of the low end also increasing very fast. As far as the high end, we don't believe the trend for the last year drop will continue. Okay. Innovation will surface to drive the momentum of the unit growth. In total, we still estimate the growth rate will be about 5% in unit growth. Silicon content I mentioned is also about equally important. That's the general model we have.
Just a question on the compute side. Could you talk a little bit about the takeoff of the revenues, given that you're expecting tape outs in first half next year? Typically, the end market that you serve in compute, especially the data center compute, takes a bit longer for qualification if you look at server market, for example. Are we anticipating revenues coming in with 7nm in 2018 itself for compute? Or could you talk about your revenue model for the compute part of the site? Thanks.
The question is regarding the high-performance computing, how would that appears to be? At this point, we think, first of all, the many of the low-end servers has. Some of our smaller customer has penetrated already, okay? But for the ballpark, the main portion of the server, I think it will wait for the 7nm to get some share by our customers. 7nm will happen in 2018 and afterwards. You'll start from 2019 and beyond. For the data center, I think it will be a more hockey stick beyond, probably really bigger portion beyond 2020. In the next 5 years, mostly are the computing of the cloud edge devices. Of course, including the networking, storage, and other cloud edge devices I just mentioned. Okay, thank you.
Next, we will have Credit Suisse, Randy Abrams, to raise the question.
Yes, thank you. I wanted to ask the first one, actually, another one on the growth drivers, just to get a feel for the base. Should we think of computing now just that computer end market that's 8% of revenue? Is that when you're thinking about computing driving a quarter of the growth, it's only 8% of revenue now. Are we thinking the same definition? Within computing, how much are you banking on Arm and IBM-based servers versus some of the other things like the deep learning, or if it's like the AR, VR graphics that you're banking on for most of that growth?
Can you repeat the question?
Yes. The question, the starting point is computing. When you are referring to computing driving a quarter of the growth, is it 8% of revenue now? We should think of your computing end market category. For the growth, are you factoring most of that growth from the Arm, IBM-based servers, or is most of it more from the VR, AR, AI type stuff?
Okay. When I talk about high computing, it does not include a lot of PC or tablets or current computing. I am more, category the high-performance computing part. Of course, data center, networking storage, but mostly is in the next three years, I think it will be mostly the machine learning, AI, augmented reality, gaming, those computing devices. Okay.
Okay.
Data center, I do not think it will come out a big volume until 2018. Yeah.
I had a two-part question on the gross margin. For third quarter, if you could go into the factors, sales are up 15%, but the gross margin is kind of flattish. If you could talk about the factors, say, from InFO or ramp of 16nm, what is driving the factor you are not seeing the leverage from Q2 to Q3 in gross margin with the higher sales? The second part is, as you ramp 10nm next year, this year you had the benefit of 16nm being a more mature process borrowing from 20nm. Would there be any margin impact on the first year of the 10nm ramp in 2017?
Randy, you refer to the margin for the third quarter. As we are continuing ramping 16nm. 16nm will account for a bigger part of our revenue in the third quarter versus the second quarter. As this year is only the second year of the 16nm ramp, it still has some dilution to our corporate average gross margin. This is a negative for third quarter.
Most of it is just 16nm and InFO.
InFO is very small quantity. The quantity comes on the line in the fourth quarter.
Okay.
C.C. was mentioning the fourth quarter will be bigger than $100 million with modest margin. It will have some dilution as well, but very small.
Okay. The second part was if any startup impact on 10 nanometer next year, since this year, 16 was second year and also helped by the 20 learning. If 10 might have an impact?
Yes, the answer is yes. We will start to mass produce 10 nanometer from the first quarter next year. For the whole year, 10 nanometers margin will be below corporate average, it will be a dilution factor for the next year, whole year.
Okay. Thank you.
Next question will be coming from Citigroup's Roland Shu.
Hi. Good afternoon, Mark, C.C., and Lora. I think the first question is for the high performance computing. Arm said for their Arm-based CPU for server, one being have a significant market share until 2020. At that time, they expect about the 25% of the share. That means for this Arm-based CPU in this high-performance computing, probably won't be very limited from now to maybe 2018. That means for the application process point of view, I think that actually will be using our 16 nanometer and even 10 nanometer because our seven nanometer, I think the one mass production from 2018. My question is there any performance gap on this 16 nanometer or 10 nanometer, in terms of this Arm-based CPU performance? That's why, for people cannot adopt Arm-based CPU on the server or even on this high performance computing quickly. Thank you.
On the Arm-based server or any way, the CPU chip was server indeed. To the main bulk of the CPU, we will have to work with our customer on seven nanometer, and is ongoing actively. Before that, I think there are customer working on 16 and even 28 to enter low-end server already also. For the other high-performance computing application, really, the entry technology is spread quite wide. Just take the ADAS, for example. 28 nanometer is prevailing entry for the ADAS technology in many of the high-end cars already. For machine learning, 16 nanometer, what do you see mostly already high volume is on 16 nanometer and entry to developer market already. The gaming is already also entry in the 16 nanometer soon. It's just started on 16 nanometer already.
For other high-performance computing, the application will happen much earlier, already on 28 and 16 and 10 also. That is different than the data center.
Okay. This high-performance computing will account for one quarter of the next five years' growth. How big it will be from this high-performance CPU for server or for this data center?
Well, that depends on how much growth we have, right? Right now, I think it's currently, because of the forecast has a big range between 5%-10%, so it is very difficult to pin down the specific number. It's about one quarter of the growth portion.
On the Arm-based CPU in notebook or PC, I think last quarter, Mark, you said you expect it will happen for maybe tablet first. Are you still seeing the same trend, and how big it will contribute to our revenue for the next five years?
I-
Yeah. I look at the transcript.
Okay.
Yes.
Tablet, of course. Application processor has been widely used in tablet already, right? It's only the tablet, the growth rate is not growing. The unit number is not growing. You can calculate how much the revenue is already. It's not a big number.
Okay. Thank you. My second question is, I look at your second quarter result. Your 65 nm revenue actually have increased, a lot. I think that actually has been 65 nm revenue have been decline continuous four quarters, and I think that will be a first quarter in the past, maybe one or two years, we see the growth. Can you elaborate why? What's the growth driver for this 65 nm increase in second quarter? Thank you.
You are asking about the 65 nm, why is the increase?
Yeah.
Oh. It's because of some major customer, they have increased their demand in automotive area. I cannot say more than that.
Will this be continued, into second half or into next year?
I would believe so, because automotive business is a very steady business, is increasing in the number of units, and we enter into mass production since the beginning of this year. That's why you see a sudden increase.
Thank you.
Okay. Next question will be coming from Morgan Stanley's Charlie Chan.
Thanks for taking my question, congratulations for great result. Very good execution. First question is regarding your smartphone content per box. Last year, you gave us some data points. I think that is super helpful. If you can quantify the absolute dollar of the contents in high-end, mid, and low-end, I think that would be very helpful. Last year, I mentioned that the high-end content is around $18-$19, mid-end $6-$6.4, low-end is around $3.4. What do you see today?
I probably cannot provide specific dollar, what I can say is the high-end content will continue to increase even into the next year. The mid-end and low-end is kind of flattish.
Okay. My next question is on CapEx spending. You mentioned that the increased CapEx for this year is actually for next year demand. Is that more for 28 nanometer, 16, or 10 nanometer spending?
It's mainly for a 10 nanometer and seven nanometer combined. We see a stronger demand than we earlier expect, and we required to build more capacity next year.
Okay. Thanks. Just a very quick clarification about the management's comments. Firstly, on the cash dividend increase. You mentioned that you will increase steady in coming years. How about next year? Is it a done deal?
I will not tell you how much. It's steadily increasing. You will know by then. Okay.
Will you start from next year? Will you start from next year?
Yeah. We have been increasing dividend quite a bit. We mean it. When we say steady increase, we mean every year, we'll consider to increase.
Okay
dividend.
Thanks. Mark mentioned that you will.
Well, Charlie, this is already going beyond.
Okay
two questions.
Sorry about that.
Sorry. Okay. The next question will be coming from Goldman Sachs, Donald Lu.
My first question is on seven EUV. I think TSMC and Mark just commented that TSMC plans to insert EUV at seven nanometer, if it is mature enough, before five nanometer. My question is, if another foundry, let's say, starts seven nanometer with EUV versus TSMC insert EUV at seven nanometer, what the implications on performance and product? For example, will TSMC's customers have to redesign the seven nanometer when you insert EUV in the second year of the technology ramp? That's my first question. The second question is on demand. What is the smartphone as a % of revenues in the third quarter? Will this be, let's say, 60% revenue or less? Also, when you saying in the next few years, smartphone would be half of the growth, would you imply smartphone and non-smartphone will grow at about the same pace?
Okay, on the first question, on EUV. Our seven nanometer will be qualified first quarter next year, so it's only three quarters from now. Definitely, EUV will not be ready. Our customer product tape-out will happen in first half next year. Definitely, we cannot put our customer at risk using EUV. This is because we are very aggressive in delivering seven nanometer, and indeed, adoption is very wide range. Our plan to aggressively using EUV will be in five nanometer, where the five nanometer will finish core in the first half of 2019. We already starting five nanometer development. EUV, at this time, we cannot use that as a volume for R&D, because doing R&D, a lot of activity going on. We choose that we aggressively get into seven nanometer without EUV, and we extensively target EUV on five nanometer.
Yes, indeed, one of our customers plans the EUV on their 7nm. The schedule-wise, it's very similar to our maybe a little bit ahead, about 5nm. A few quarters. Our 5nm is two steps down. The density is very aggressive using EUV. Learning after 7nm will be very safe on the 5nm. Had the EUV available before 2020, we consider using our 7nm technology as a base to adopt EUV for the probably the second wave products. We want to minimize the design changes so that all the customers coming to our 7nm will be easily adopt what we can reap from the EUV on 7nm. That is our plan. I think we develop our technology at the best timing for our customers' products. That is our current plan.
The second question is on smartphone as a % of revenue in the third quarter after Apple ramp. Also, would you imply smartphone, non-smartphone will grow at the same-
I don't know about the third quarter, generally, currently, smartphone is about 55% of the corporate revenue. Wafer revenue. Okay. Yes, if the growth in the next 5 years, smartphone related will be half of that.
Related to this, what about earnings volatility? For 16nm or especially 10nm, your largest customer will be essentially dominating this particular node. If this customer has seasonal or inventory issues, how can TSMC fill the leading-edge capacity with very limited number of customers?
I think, first of all, we are everyone's foundry. Our customer covers all areas of the application. We do have big customers. That is not new. I think your question is the volatility of the big customers.
Yeah.
Getting-
For 10 nanometer, let's say your largest customer take 80% of the capacity, and if he has two seasonal or inventory issues, then you.
I see.
can feel it.
That's why we design our 7nm equipment-wise is 95% convertible. Ballpark is, you can consider investment-wise, 10nm and 7nm can be converted easily. That's how we minimize the volatility of the 10nm.
Next question will come from Deutsche Bank, Michael Chou.
Hey, Lora. One question is, do you expect the ASP increase for this year?
ASP?
You're asking ASP?
Yeah. ASP increase for this year, for whole year.
You're talking about the blended ASP?
Blended ASP.
Yeah. I think so.
Yes. We expect blended ASP will go up this year.
Would that be low single digit or?
I prefer not to quantify that.
Thanks. Second question. C.C., you mentioned the second generation of InFO for next year product. Would that be for seven nanometer product or 10 nanometer?
Start with 10.
Starting with 10, right. It will start to have revenue contribution next year, 2nd stage?
It should have.
Okay. The first half or second half?
Second half.
Second half.
Probably, more probably.
Thank you.
All right. I think we will start questions that has been queuing in the line for quite some time. Operator, could you please have the first caller on the line?
Sure. Our first question comes from the line of Brett Simpson, Arete Research. Please ask the question.
Thanks very much. With regards to InFO and the CoWoS, as I see it, you're moving from selling wafers to selling turnkey chips, only good die, which is a big change in your business model. If you look out over the next few years when you ramp seven nanometer, what portion of your leading-edge sales might be adopting this new turnkey model? And when it comes to DRAM memory and package, does TSMC take responsibility for sourcing and holding inventory for memory, or do you expect your customers to fix this? Thank you.
Okay, let me repeat Brett's question. He's asking, how much of our seven nanometer business will also be using either InFO or CoWoS that TSMC is providing a turnkey solution. What's the proportion of our seven nanometer business that will be of this type of business model? The second part is about memory in the packaging solution. If memory has any problem, whether it will be TSMC or the customer who is responsible for that problem.
Well, let me answer the seven nanometers at how much of the percentage will use CoWoS and the InFO together with as a CoWoS.
Yeah. Just turnkey solution.
Just turnkey solution. Probably quite a big portion because of, we offer InFO for a very cost-effective packaging service. That will help to improve the total packaging efficiency in terms of chip area or in terms of thickness. For CoWoS, it will help to seven nanometers high computation space to competing a very high-performance area. That will be a smaller percentage, but we see the potential. That's what I just mentioned in my presentation. Put all together, I would expect quite a significant amount of seven nanometers of product will be using either CoWoS or InFO. That's one thing. Talking about the memory, stacking together with seven nanometers chips. Certainly, we offer the turnkey solution. We are very flexible in working with the customer that all the possible combination are all possible.
Does that mean that you would purchase the memory and book the memory sale within the overall package? Would you leave your customers to purchase the memory and take inventory risk?
Well, let me see if I hear you correctly. You are asking whether or not we are paying for the memory and then charge the customer as part of the revenue that we obtain from InFO, or it is separated?
Exactly. Yes.
Okay.
Okay. We don't buy the memory per se. We work with a memory supplier so that their spec in terms of mechanical stress, in terms of a lot of things, those kind of things. We work with them, but we don't buy memory and resell to the customer. However, we are responsible to stacking them together and as a whole packaging to our customer.
Thank you. Maybe just one final question. When you look at TSMC 7nm for high performance compute, can you give us your perspective? How would you typically compare this to Intel's 10nm, which will be in production relatively similar timescales? Yeah, thank you.
Can you repeat the question please?
Yeah. How do we compare our seven nanometers performance and whatever the characteristics versus Intel's 10 nanometer, which Brett believes that timeframe-wise, we are about to offering these two technologies about the same time?
Well, you have to ask our customers. I cannot speak for them. They definitely have their plan, and we deliver our technology. Of course, as you know, they deliver their architecture differentiations. Really, the end product performance depends on multiple factors. I cannot speak for my customers at this point on this.
Okay. Let's move to the next caller on the line. Operator, please.
Sure. Our next question comes from the line of Steven Pelayo from HSBC. Please ask the question.
Yeah. First question on 20 nanometer. It's still your largest segment, but it did decline last year, yet the total company still managed to grow about 10% or more in 2015. What do you think for 20 nanometers this year? What type of growth, I assume you think it's going to grow this year, are you expecting?
Steven, we really cannot hear you that well. Can you sort of move a little bit away from the microphone? You're asking about revenue contribution from different geometries.
Yes. Maybe you can hear me a little better now.
Yeah, please.
No, I was asking specific to 28 nanometer node.
28.
It's still your largest node, and yet it declined last year. Though I was asking for this year, do you expect it to grow and by how much?
Okay. Whether or not we will grow 20 nanometer, and by how much in this year?
Correct.
Well, Steven, in fact, we deliberately combined 16nm with 20nm, and we report that as one node, so we do not separate that. Sorry.
I was talking about specifically 28nm.
Oh, 28nm.
That's your largest node.
I see.
Do you expect that to grow?
Sorry, 28.
Yeah.
28nm. Do we grow 28nm, and by how much this year?
Correct.
Well, while Lora is checking with the number, all I can say is our revenue this year is slightly lower than last year. Slightly. Percentage-wise, because our revenue increased, the percentage of the 28 nanometers of revenue
C.C., actually, according to our forecast, 28 nanometer dollar-wise will be higher this year than last year.
Will be higher.
Yes.
Okay, good. Okay, as I said, stronger demand in the second half.
I confirm that. Okay.
I apologize.
Okay, fine. That's fine. The reason why I'm asking this question is really to talk a little bit about above 28 nanometer. Competitors like SMIC are growing very fast this year and their capacity constrained. They don't really have any 28 nanometer really to speak of. What's going on with the above 28 nanometer nodes? Above 28 nanometer, it seems like maybe TSMC is losing share. Is that right? Is that just China influences? How would you respond to that question?
All right. His question is for bigger nodes, bigger than 28 nanometer, how does TSMC what's our strategy compared to other foundries who are growing shares in those bigger nodes?
Okay. For the older generations, our strategy is simple. We don't particularly to increase the capacity.
we develop some derivative technologies, such as in the CMOS image sensor, embedded flash, power management IC. To keep the whole capacity fully loaded, that's our strategy right now. We don't substantially increase the capacity as other foundry peers doing.
maintain good profitability.
Okay. That's understood.
Thank you. Okay.
That's fair. My last question, if I could just sneak in here, is about the dividend. You guys are raising your dividend, yet you're generating more free cash flow every year, so you're growing your cash balance. I'm curious, when does this become too much? When does it start to impact your goal of sustaining greater than 20% ROEs? Is there an optimal cash level and an optimal capital structure that we could talk about instead of focusing on a dividend payout ratio?
Steven's question is, we have been growing free cash flow every year, and our cash balance continue to go up. What would be a capital structure of TSMC or payout ratio?
Steven, we don't use payout ratio. Our dividend policy has rooted as sustainability. We have started to pay dividend since 2008. Since 2004, actually. There is a very long period of time. We paid TWD 3 every year for consecutive 8 years. Then we significantly increased the dividend to TWD 4.50 a year ago, and then to TWD 6 today. If I look at from today's forecast, we have confidence we can continue to grow our free cash flow every year. That is why I say the dividend policy will kind of change from sustainable to sustainable and increasing. That's the thinking that we have in mind.
Is the current TWD 668 billion in cash, is that an excess amount that you need? What is an optimal cash level to run your business? That's my last question.
The optimal cash position depends on what you do with your cash, okay. I think for now, our cash is mainly to support our organic growth. We may have some M&A, but we don't know at this moment. We'll try to be a little bit conservative in keeping enough cash to do the both. One way to support a sustainable increase in dividend. On the other hand, to have some buffer if there's M&A case.
All right. Operator, let's move on to the next caller on the line, please.
Sure. Our next question comes from the line, Mehdi Hosseini from SIG. Please ask the question.
Yes. Thanks for taking my question. Can you hear me, Elizabeth?
Yes, we can.
Thank you. I have two questions, one on CapEx. A couple of years ago, when you first started investing 20nm, there was a surge in CapEx as you were preparing 20nm. Now it seems like 10nm and 7nm are pretty much comparable to 20nm and 16nm. Back then, a couple of years ago, after significant spending in 20nm, there was a period where your CapEx came in flat, or even last year it declined. Should we assume the same kind of a CapEx spending looking forward, especially since 10nm and 7nm are comparable to 20nm and 16nm?
Can you repeat the question, please?
Right. Mehdi's question is, when last time he saw TSMC having a surge in CapEx, that was the time when we built 20nm, when we ramped 20nm. He's asking whether same type of CapEx pattern will be repeated in the future when we ramp, say, for example, 10nm.
We don't believe so, because as Mark was mentioning, there's a very high % of conversion rate. When we roll to the next generation of technology, a big part of it will be converted from the previous generation. You will not see a step up type of things that we have experienced in the past. It will be gradually increasing. Mm-hmm.
Right. A step-up is happening in 2016. In 2017, your CapEx may not increase. Actually, it may be flat or down as the conversion starts to take place. Is that correct?
Whether 2017 CapEx will be flat or down compared to this year because of conversion.
It will not. As I just said earlier, I think you will not see a step-up kind of things. We'll look at the business every year, and depending on how much leading-edge capacity we want to build in that year, also depending on the profile of the ramping. That also another factor that will affect our CapEx. It's probably better to follow my earlier guidance. We believe in the next 5 years, our CapEx intensity will be in the range of mid-30s.
Okay, now.
I have a follow-up question regarding the near-term trend. You said that the customer days of inventory is going to increase in Q3, and you're not sure about the sell-through. Is it going to take a couple of more months to better determine if customers are overbuilding, or do we have to wait till January, February timeframe? How do you assess this? Because your commentary is a little bit confusing, especially regarding days of inventory.
All right. Mehdi's question is, how do we know whether or not the inventory that we are seeing today will be too much? Do we have to wait until next January or February to know, or do we have a method to assess?
Indeed, we cannot see crystal clear about sell-through. What we have tell you about the unit growth is really the sell-in to the phone service providers. Sell-through, indeed, there are uncertainties. We assume the growth of 6% increase, and the inventory level we cited is assuming the sell-through is also a 6% increase. If it differs, the inventory level will also differ accordingly.
Thank you.
All right. Thank you, Mehdi. Let's come back to the floor. The next question will be from Credit Lyonnais, Sebastian Hou.
Thanks for taking my questions. My first question is to follow up on the back end, the advanced back-end technology. C.C. guided this year, first quarter revenue of InFO will be above at least TWD 100 million. Can you give us some guidance for fourth quarter next year? Also the first quarter 2018, when InFO and CoWoS combined?
The fourth quarter this year is high volume ramping. Following next year is supposed to continue this kind of production, I believe. Next year, the whole year will be much better than just this quarter times four. You are talking about in 2018, CoWoS and InFO together or 2017?
2018.
2018. I would believe CoWoS has a high potential, but it's still a small volume as compared with InFO. InFO is much more high volume production. Most of the revenue will come from InFO.
Okay. Based on your first quarter 2016 guidance, it will probably account for 1%-2% of your 4Q revenue for InFO. Is there any, in terms of revenue contribution percentage for 2018, in terms of InFO and CoWoS combined? Would it be higher this 1%-2% run rate?
I would hope so, but I don't know the 2018's revenue yet.
Okay. Thank you. My second question is on the 7nm and 10nm. Six months ago, I remember I asked the questions and Mark Liu and C.C. Wei answered me that the 10nm revenue contribution in 4Q next year will be higher than the revenue contribution of 20nm back in fourth quarter 2014. I wonder, does that still hold?
Elizabeth Sun, can you repeat the question?
Okay
quite get it.
Sebastian Ho's question is, compared to 2014 fourth quarter revenue from 20nm with 2017 fourth quarter revenue from 10nm, which one will be bigger?
Right now, the dollar-wise, we think it's bigger, okay? The percentage-wise is next year's business. Those numbers are too early to predict at this point. Yeah. This 10 nanometer is very strong, and we cannot give you a specific number for that. It will be around similar level, I think.
I think 4Q 2014, our 20 nanometer was 21%.
Yes.
2014 compared to 2017 will be a smaller year.
Yeah
overall-wise.
Okay. Understand. Given that you have more tape outs and design tape outs on the seven nanometers compared to 10 nanometers. We can assume that by fourth quarter 2018, the seven-nanometer revenue contribution, or in dollar amounts, should be a lot higher than 10 nanometer contribution fourth quarter 2017.
This is 2018, and we are middle of 2016.
Yeah.
Sebastian.
But-
Whatever the guidance we can give you may not be reliable.
You guys have a very long visibility, I thought so.
Not that long.
Okay.
Not two and a half years.
I think we're going through ramp, 10 nanometer going through ramp next year, and 7 nanometer and 10 nanometer will continue the ramp in 2018. Okay?
Okay.
Thank you.
Thanks.
Next question will be a follow-up from Morgan Stanley's Charlie Chan.
Thanks for taking my question again. I just want to clarify Mark's comments on smartphone market share. You mentioned that you believe you can continue to gain market share in the smartphone semiconductors, but now you essentially have a very high market share now, except for those in-house production from Samsung. Where is the upside for you share market share in smartphone?
Well, we and our customer work together to increase their shares. Yes, currently our smartphone with our customer, market share is already very high. We will gradually increase. We think we'll gradually increase further. Okay. Many of our customer today in mid-end were also attempting to get into the high-end, and particularly the silicon content of high-end and mid-end also increasing, which favors leading-edge technology, what we provide. Those are the advantages that we have with our customer to be able to get increase the market share we have.
Okay.
Okay.
That's very clear. Gain market share along with your customers.
Okay. I just want to follow Donald Lu's question on this short-term volatility, right? You raised CapEx for next year 10 nanometer demand. I think the fact is that, you disclose in your 20F, right? Apple and Qualcomm, each accounts for 15% of your 2015 revenue, and we all know that Qualcomm may walk away to Samsung temporarily. For Apple, you also mentioned that you cannot predict the sales of, let's say, iPhone 7. How can you be that confident to spend the CapEx for 2017 10 nanometer demand? Thanks.
We cannot give you a specific number. It doesn't mean we are not confident. I think, of course, we want to support our customer for any growth opportunities this can achieve. At the same time, we manage carefully about our capacity build-up. This is a very complicated discussion with our customers. At this point, I think those capacity we built will be needed by our customers.
Is there any early indicator you will win, keep very high market share as a 10 nanometer project for big customer?
Yes. We have been working with customer for a long time for 10 nanometer. We believe our market share will be more than 70% next year.
Okay, thanks.
Okay. Follow-up questions from Citigroup's Roland Shu.
Thanks. Lora, I just take your assumption for 16 nanometer gross margin to be about one percentage point dilution of the corporate average in second quarter and Q3. I just did a calculation. It seems in Q3, actually we don't have any benefit for the higher utilization on gross margin point of view. Does that mean that in Q3, the average 16% revenue growth is mainly from the product mix change? Should we evacuate now?
The hotel is announcing a fire on the 29th floor.
Do we have time now?
Oh, okay. [Foreign language] Are you sure? Yeah. Okay. It's not a real fire. It's just a drill. It's a fire drill taking place at this time, so we will continue.
Okay.
Yeah. Roland, please.
Yeah. I think TSMC's practice is when you have this alarm everybody need to evacuate. Yeah. Yeah. Okay, I think my question, just repeat my question. I just take your assumption, 16 nanometer gross margin to be one percentage point dilution of the corporate average in second quarter and Q3. My calculation shows in Q3, even though we have about 16% revenue growth, however, it seems we don't have any gross margin benefit on the higher utilization. Does that mean that this 16% revenue increase is mainly from the product mix change? In Q3, we have higher depreciation, to have a drag of the gross margin? Thank you.
Basically, Roland, your question is whether or not we will have a higher utilization rate or a lower utilization rate in the third quarter compared to the second quarter, and that's the reason for not giving you a 55% gross margin, right?
Yeah. If the depreciation is much higher than second quarter.
Higher depreciation, yes.
It's a combination of many factors. We do have higher depreciation. Just imagine we have a back and loaded CapEx, it's naturally depreciation go up from second half. That's one thing. Also we are ramping 16 nanometer, which we still have some dilutions, and we have a better utilization in second quarter as well. If you took all those three factors together, that's the guidance that was provided, grounded based on those.
Yeah. Can we have a rough number for how high the depreciation in Q3 it will be?
I think it's TWD 2.5 billion more.
Yeah, it's about TWD 2.5 billion quarterly depreciation.
It's about a one percentage point lower in gross margin.
Roughly.
Yeah.
Yeah. Okay. The whole year depreciation, is still increased about mid to high single digit. Is that the same?
It's in the range of mid-teens year-over-year. Okay. Mid-teens. Yeah.
Mid-teens.
Mid-teens. Yeah.
This is very different from the number you talked-
No, I was talking about the same thing. Yeah. Depreciation year-over-year will increase by mid-single digit. Yeah. I didn't change that.
Okay. No? I think it's not
Low single digit.
Mid to high single digit.
Low single digit.
Oh, I'm sorry. I was looking at wrong line. Yeah. Mid-single digit. You're right.
Okay. Thank you. Thank you for the clarification. The last question for the DOI inventory. Twice, you said in Q3, customers still restocking, in Q3 and the Q4 will be stocking, and the inventory days were back to seasonal level. Does that mean that from first quarter next year, are we going to see a normal seasonality, normal quarter in first quarter, or customer are going to restocking again from first quarter next year? Thank you.
I think that there's still uncertainties depending on how the inventory situation in the fourth quarter and sentiment over sentiment in the fourth quarter. It's a little bit too early to say that.
Okay. Follow-up question from Credit Suisse, Randy Abrams.
Thank you. Just to follow up on Roland's question, to pull it back to fourth quarter, where you talked about the inventory and the high base, is there a way to think about it relative to seasonal? Like, what's the seasonal fourth quarter, and then if you're thinking it's normal or below seasonal? Yeah, question is fourth quarter, like, what you'd expect normal seasonal, because you mentioned it's normally a correction quarter. What's the normal magnitude and then how you see At this stage, it could change, but at this stage, how you see it relative to seasonal?
Fourth quarter is a seasonally will correct. It's a will correct. What we say from third quarter, little bit higher than seasonal, going to fourth quarter seasonal, including both factors, correcting the deviation of seasonal and also between third quarter and fourth quarter, seasonal will be corrected. The days will decrease. Yeah.
Okay, do you think the decline Because it was above seasonal in third quarter and inventory above, we could see below seasonal fourth quarter. Do you expect you could see a below seasonal fourth quarter because.
No, no. We see it's close to seasonal.
Okay. My follow-up question, IoT is one of your growth drivers. Your competitors like Samsung, GlobalFoundries have been pitching FDSOI. Do you see any merits to that track, or how do you see customer interest in that relative to the strategy you're adopting and any need to consider that in the future?
Well, let me answer that question. For the FDSOI, the main merit is kind of low power consumption Which I think TSMC has offered a complete set of technology roadmap for the low power applications. We started from 14 nanometers low power, then entered into 28nm. Now even 16nm, in fact, we offer down to 0.5 volt operation voltage. According to our experience, we working with our customer, the roadmap has been very satisfactory to their product needs. We expect we are very well positioned to capture all the opportunities.
Let me add to that. I think FDSOI may be useful for specific product, but our bulk CMOS ultra-low power has a much bigger design ecosystem. IoT is a very fragmented designs, and many, many customers are trying to enter that. We think the ecosystem of bulk CMOS is much, much useful for our customer than the niche FDSOI application.
Thank you.
Okay. There is a question in the back. I think you are Merrill Lynch sales, right? No, you are not. Oh, UBS, sorry. The lady in the back is sales from UBS.
Thank you, Dr. Sun. This is Julie from UBS. My question will be actually more simple than the analyst. I am actually quite interested regarding to the market share comment that Mark has made. I am just wondering, regarding to the market share gain, would that be coming in from, let's say, your existing clients making into higher end of smartphone, tapping into area they never happened before? That is one thing. That is called market share gain, right? Secondly, is it existing client coming back to you? That is also a market share gain. Which one would that be?
Both.
Okay.
Yes.
Which one is happening faster?
It happened at the same time, yeah.
Okay. Would you be able to quantify the market share of TSMC in smartphone?
Currently, we are about 55%, and we expect to increase from that number further up gradually, yeah.
Okay. My second question is actually, what's TSMC's view on ASML buying Hermes? Does that help you, going forward, in terms of tools procurement, or could you comment on that?
HMI, Hermes, has this same defect inspection tool. It can detect very fine feature on wafers. ASML and HMI merger intent to be able to develop the lithography more precisely control according to the wafer images. I can only speak from industry or technical perspective, that definitely will help the capability of a lithography tools, and therefore help the technology capability, and help the technology development for their customers. That's my comments, yeah.
Well, all right. Last question coming from Credit Lyonnais, Sebastian Ho.
Thank you, Dr. Sun. Just two quick follow-ups. First one is, we noticed the consumer application in second quarter increased significantly, both from quarter-on-quarter and year-on-year perspective. Can C.C. or Mark comment on that?
Video games.
Okay. Thank you. A very quick one. The last one is on the CapEx increase, about $500 million. I wonder whether that suggests some pull in of the 10 nanometers expansion.
Okay.
Yeah. Yes. It is. It's for 10 nanometer, because we need a bigger capacity next year.
He's asking whether it's a pull in, which is a schedule issue.
For next year, it's not. It's just higher demand, we need more capacity.
That doesn't have any correlation with the mass production schedule?
Schedule stays the same.
Same. Okay. A follow-up on that one is, does the CapEx mix still about 10% go to back end in terms of the new CapEx? Or they were skewed more toward the advanced technology, focus on 10 nanometers and 7 nanometers for this year?
The back end CapEx for next year probably will be much smaller than this year because we build up InFO capacity already. Next year, we are going to be the second generation, probably just improve some features. That's all.
Thank you.
Okay, I think it's about time. Thank you very much for coming to our quarterly conference. I hope you will join us next quarter. Our transcript will be available very shortly, as well as the video replay. Thank you.