← 返回

2026.09.23 | TB | Semis | Sony & TSMC Image Sensor JV

原标题：Sony & TSMC – Imaging Sensor JV & Takeaways From JASM

#semis

Third Bridge · 2026-09-23

Analyst-led • 8 Sep 2026

Companies: 6758 JP, 2330 TW, 7203 JP

Yuan Kuei Huang (YKH) Senior Data Scientist at Texas Instruments Inc

Michelle Lee (ML) — Third Bridge Sector Analyst

Agenda

Key Insights

Introduction

ML [00:00:03]: Welcome to Third Bridge's Interview entitled Sony & TSMC – Imaging Sensor JV & Takeaways From JASM. I'm Michelle Lee, and I'm joined today with Yuan Kuei Huang, Former Principal Manufacturing Engineer at TSMC.

Before we start today's Interview, please state I agree or I disagree to the following statement: You understand the definition of material non-public information and agree not to disclose any such information, or any other information which is confidential, during this Interview.

YKH [00:00:42]: I agree.

ML [00:00:44]: Could you please begin with a brief introduction to your background?

YKH [00:00:54]: I've been working nearly 13 years in TSMC. Currently, I'm working in Texas Instruments. In TSMC, I majorly focused in the equipment automation and resource reduction. For the first nine years, I stayed in Taiwan and focused on these topics, and the last four years, first, I went to Tsukuba in Ibaraki. The site there, I focus in… The department is the 3DIC packaging R&D centre. Last year, I transferred to Kumamoto and joined JASM, which is the joint venture company of TSMC and Sony. There I focused on mass production and also coordinated with a lot of Japanese engineers, and we plan to build a new company there. I think this is the basic introduction.

Questions

[00:02:37] What were the primary operational bottlenecks or execution risks during the JASM [Japan Advanced Semiconductor Manufacturing] fab's ramp-up?

YKH [00:02:59]: TSMC has already many mass production ramp-up experience in Taiwan, but when it comes to JASM, the first one is that the vendor, total difference. I think that the manufacturing segment in Kagoshima, they suffered a lot of culture difference with them. The way how to build the structure of the fab and the rules, they are totally different. They spent a lot of time and extra money to do negotiation. The second one is that the engineers in Japan, because for nearly 30 years, the Japanese students, they no longer have interest to being semiconductor engineers. That is hard for them to catch up the latest technology or skills, so we trained them from zero. Also, actually the location is in Kumamoto. It's not very attractive for students in Tokyo University. Even you pay a relative good compensation for them, but maybe still not interested.

ML [00:04:55]: Is that still the same problem for JASM Fab 2? It's still under construction, but has it learned something from Fab 1 to become much better now, or is it still the same situation?

YKH [00:05:15]: For the construction part, I think it's better now, way much better now towards… In a better (inaudible 05.25). For the talent acquisition, I think there is still a little bit problem because the compensation, it's slightly higher than average compensation, but compared to the high-level talent we want to acquire, it's still a little bit hard.

[00:05:52] The new Sony-TSMC [Taiwan Semiconductor Manufacturing Co] JV involves image sensors. What are the most common operational bottlenecks when optimising a new front-end wafer fabrication process for CIS [CMOS (complementary metal oxide semiconductor) image sensor]?

YKH [00:06:10]: When I was in JASM, we also built some of the CIS component for them. The advantage for TSMC is that they have a very good system to do mass production. For this, actually, the know-how is controlled based on other vendors, but Sony designs for the chip. In general, even they build packaging for the chip. TSMC can build it very fast, but they don't have knowledge to design to build the CIS from zero to the packaging. For TSMC, I think the bottleneck is here that… Actually, they never did it before. They're still finding the path to do it. It depends on how much knowledge that Sony wants to share, I think.

[00:07:34] What are the differences in process steps or equipment requirements for CMOS vs logic chips? Is the former more capital-intensive or requires more process steps and other special materials?

YKH [00:08:05]: Actually, the process step is slightly fewer than the logic chip. I think, yes, the equipment is much more capital-intensive because… I think it's a different way, because in Taiwan, TSMC also do the packaging, but their packaging is a high-level packaging. For cell phone, cell phone chips or the data centre chips, but in CIS chip, the technical level is not as hard as they did in Taiwan. Actually, I think now that, same to the previous question, because the equipment that TSMC use for packaging in Taiwan, they ask local vendors, all Taiwanese vendors, to provide the equipment. In the image sensor domain, they don't have any experience and they have to follow the rule and the principle provided by Sony.

[00:09:38] How long does it take to transition a new factory from the manual engineering and ramp-up phase to fully automated operations?

YKH [00:09:53]: In terms of construction, actually, we don't have a human phase anymore. They already started from the automation phase. About transition… Do you know what is AMHS, the automatic material handling system? When the system is deployed into the fab, so the technicians don't have to move the chip by themselves and put it on the carts and (inaudible 10.30) to each station, the robots would do that. Actually, when that moment started, so they can use algorithms to optimise the delivery of the chip and control the speed, the priority for that. We already don't need human (inaudible 11.08) humans to handle the abnormal type or situations…

[00:11:18] How long does it take for a factory to enter full operation and mass production? How is the new JV preparing for a 2029 mass production target? Do you think it's feasible?

YKH [00:11:41]: Yes, I think particularly feasible. Mass production maybe in the building and moving, I think they need 1.5-2 years. They have to start the pilot run. Pilot run normally, at least, three months to half year, and they start to do the mass production, so I think the end of 2029 is achievable.

[00:12:32] In Kumamoto, from a daily manufacturing standpoint, what are the logistical challenges of operating a front-end TSMC fab adjacent to Sony's back-end packaging facilities? Is it too messy?

YKH [00:12:54]: No. I think that they divided the track very clearly. TSMC only do front-end process and Sony only do the back-end packaging.

ML [00:13:12]: What is the advantage of having two factories together at the same site?

YKH [00:13:22]: To shorten the delivery time, I think that's the major point. Because JASM, they manufacture the product in Taiwan in Taichung and every chip they have to send back to Sony for packaging, and from Sony, send to China or Vietnam to assembly. After TSMC, they just send to the neighbour.

[00:14:03] What percentage of cost do you think TSMC and Sony can save by operating in the same place vs from their separate facilities in Taiwan and Japan?

YKH [00:14:25]: Because they are building old technology, the N28 or N22 technology, in this domain. Actually, gross profit is not so high, so the transportation fee portion is a little bit higher and I think they can shorten about 30% of the cost.

[00:15:01] What operational experiences from JASM – or the previous JV – can be directly applied to the new JV? Both JVs are manufacturing different things, so there aren't many things that you can apply from the previous JV to the new one.

YKH [00:15:32]: I think for the mass production for manufacturing part, I think around 70% experience can be gained from JASM because you already had some trained engineers and you already deployed servers and systems. The issue is that they don't know what is the flow of this new CIS chip. They are still exploring the path, so maybe some of the existing systems and principles have to modify (inaudible 16.23) the unknown parts.

[00:16:30] You said there are some equipment and specification differences in CIS chips. Has the core equipment for the new JV already been fully finalised? Are there any supply chain factors that could impact or delay delivery timelines?

YKH [00:17:08]: I think the core equipment is already finalised because when they decide to build a factory, they have to know it and they have to calculate the capacity of it and not only the specification and also the quantity, I think they already calculated. Supply chain factors, I think the supply chain factors (inaudible 17.40) order from customers, how much order (inaudible 17.48) how much order they want to order.

[00:18:03] In the semiconductor industry, especially in Asia, there's a trend of localising equipment or materials if possible. Are you seeing similar trends for the new JV, or are the companies picking up any equipment or material that's good enough?

YKH [00:18:31]: The material, that is the strength of Japan, so even if they're in Taiwan, they still use the Japan materials like slurry or the chemicals. For the equipment, yes, this is obvious because Taiwan, they also want to use vendors in Taiwan and it's easier for them to control the supply chain. In Japan, yes, they were requested to prioritise the Japanese vendors.

ML [00:19:12]: Is that transferable because the core technology is Taiwanese, so you don't need to stick to the Taiwanese equipment material?

YKH [00:19:27]: Sometimes it's not, but they will be forced to do some research too. From the TSMC's angle, they also want to have a second source. Before the new company, I think that's the second source. After the new company formed, I think they might… If the quality is comparable, they might be the first choice for them.

[00:19:59] Why do you think Sony decided to form a new JV rather than continuing to purchase chips from TSMC fabs, despite the massive upfront investment?

YKH [00:20:32]: When I was in JASM, there's a huge mass production efficiency difference between Sony and TSMC. TSMC is so good in this domain. They may want to outsource the manufacturing part, but not the knowledge to other companies. They also still want to control it, so I think they made the joint venture.

ML [00:21:07]: Do you think it's worthwhile for Sony to do that, given that JPY 1tn is quite a big amount? Does the company want to control the supply from TSMC, and is that the main reason?

YKH [00:21:29]: Yes, I think that's the main reason. Also, they want to have the sponsorship from Japanese government.

[00:21:48] TSMC has an 86.5% share in the JASM JV, but in the new JV, the company only has a minority stake of 32%. Do you expect any differences in the operational practices or management styles of those two JVs due to ownership differences?

YKH [00:22:26]: As I heard, I think the portion might be adjusted. The new company, the operation level is fully controlled by TSMC because of the efficiency difference. On the management side, that is the Sony side.

[00:23:22] Could you outline any differences in management styles at Sony?

YKH [00:23:35]: I talked with a lot of managers came from Sony. The culture is totally different. For example, they don't take rest in the lunch time. The employees from Sony, when there's a lunch time, they just eat and go back to work. The culture from Taiwan, they're different. This is their rest time. They eat lunch and they rest within the hour. They even take a nap in their lunch time and they go back to work. Also, from the Japanese side, they have a lot of meetings, so many meetings day in, day out. In TSMC, there are still lots of meetings, but cannot compare to Sony. They want to spend time to do some real stuff.

ML [00:24:45]: Do you think this new JV has more Japanese characteristics?

YKH [00:24:53]: Yes.

ML [00:25:06]: What would be the differences between the JASM and the new JV? Do you think efficiencies or outcomes should differ, or are the differences only at the cultural level?

YKH [00:25:33]: I think not only culture, because the instructions, the direction from top level managers is different and the way they do their work is totally different. Japanese like to plan everything before they do it, so they have so many meetings to do that, but Taiwanese don't do it this way. They have a small meeting and do it first and check the result fast and modify the plan continuously. That's the difference between the management culture.

[00:26:39] How might Sony's majority ownership in the new JV change IP ownership agreements compared to the JASM JV?

YKH [00:26:55]: IP ownership, yes, because TSMC don't do the design or this product-level IP ownership. They have their own IP ownership for manufacturing processing, so yes, there is total difference.

ML [00:27:15]: What could be the differences between the JASM and new JVs in terms of dealing with the IP?

YKH [00:27:39]: I think there's a huge difference because the IP that TSMC cares about is the semiconductor manufacturing level, but for that specific domain like CIS, they normally don't care about it. Also, this company is majorly controlled by Sony. I think that's part of the reason that Sony wants to keep the IP on their hand.

[00:28:14] How will Sony protect its proprietary pixel layer IP while operating on shared fabrication tools with third-party foundries?

YKH [00:28:37]: Actually, I'm not familiar with this kind of domain, but I think TSMC (inaudible 28.46) other customers, they have comprehensive protection from the customers' IP and TSMC has no interest to compete with them, so yes, they just treat them like a normal customer is enough. No need to do extra protections.

ML [00:29:17]: Isn't there any system or method that Sony can use to protect its IP in these cases, because TSMC or other foundries are not interested in this and are not competing with each other?

YKH [00:29:40]: Sorry, I'm not familiar with the IP protection.

[00:29:51] JASM brought in Denso and Toyota. Is it likely that the new JV will eventually dilute equity to other companies, such as automotive or mobile customers, to secure demand in the future, similar to what JASM did?

YKH [00:30:31]: I think only Sony joined the AVSM binding, because this site is especially for CIS components. I think Denso and Toyota is not that interested in this area. In JASM, they also do some chips for automotive, so Toyota joined the joint venture, but they don't speak. They just provide the money.

ML [00:31:08]: Is the image sensor JV also really important to automotives?

YKH [00:31:18]: Yes, it's important, but they don't have to join the investment. They can directly purchase from Sony.

[00:31:32] Would it be better for Sony and TSMC to dilute their equity to customers because it decreases investment burden, or should the companies keep all the shares and sell chips to customers?

YKH [00:32:01]: For the CIS domain, I think it is better to… From the angle of Denso and Toyota, they don't need to join their investment because actually, they have second source like Samsung and being a customer maybe is still (inaudible 32.43) to purchase the requirement they need because if they join the investment, they have some responsibility to put all in here and maybe some effect maybe to do it, in choosing those.

[00:33:12] How would you evaluate the potential for physical AI applications – which are the current theme in the market – to shift the demand landscape for CIS in the next 3-5 years?

YKH [00:33:55]: Yes, I think it's definitely yes. You can see, they made Cybercab. This is robotic, anything like car or system, android. They highly depend on the visualisation. I think CIS will take a very, very important portion of this in this domain section.

[00:34:29] What percentage of CAPEX is expected to be offset by the Japan government's subsidies? You mentioned that facilities are being set up partially due to government subsidies.

YKH [00:34:55]: The Japanese government invested JPY 60bn. Yes, it sounds a lot, but for the semiconductor factory building, I think this is only maybe less than around 10% only. I think it's very small portion, that doesn't help us.

[00:35:23] Do you think government subsidies come with any restrictive operational terms or conditions, such as extra documents or mandatory shipments to certain Japan-based companies?

YKH [00:36:04]: I don't know the exact term they stated in the contract with government, but yes, we can feel it. You have to hire localised engineer first. As we discussed before, you have to purchase the raw material from local first and then the vendors from Japan first. I think that they want to bring in the whole industry. I don't know what is the detail, but we can feel it when we work there.

ML [00:37:02]: Do you think that undefined condition is less efficient than your usual practice? Sometimes using a local Japanese vendor first would be a little bit more difficult than using TSMC's traditional, existing vendors. Have you seen any differences between those two aspects?

YKH [00:37:42]: Yes. Equipment, we use a lot of vendors from like Applied Materials, KLA, but if TEL provides the same, the Tokyo Electron provides the same equipment, we were requested to change to that. That might be different than the process that they invented in Taiwan. We take extra time for that and also for the human resource that, actually, Japanese students, they don't have interest in semiconductor manufacturing.

[00:38:37] Looking at the current image sensor market share or technology trends and pricing, which specific end user markets are the JASM and image sensor JVs targeting?

YKH [00:39:27]: For JASM, at first, they focused on the automotive and CIS. I think after this company, the whole CIS part will move to the new company. They will be majorly focused in the CIS part because that's the strength of Sony. Then, for the automotive, I think they will still in JASM because in Taiwan, they are shrinking the production line of N28 technology. Their focus will move to JASM.

[00:40:27] What are the most significant cost differences when setting up a CIS fab compared to a traditional logic fab?

YKH [00:40:54]: Cost difference, I think compared to normal like factories in Taiwan or in JASM?

ML [00:41:05]: Yes. JASM would be helpful, but you can also talk about Taiwan.

YKH [00:41:21]: I think compared to JASM, I think it's similar. Only difference is the process difference. I think that the major investment amount is similar, but compared to Taiwan, the cost in Taiwan will be much lower because for vendor selection, you don't have restrictions. You can select every vendor, even a China vendor, you can do that. You will be much cost effective in Taiwan.

ML [00:42:14]: If the cost in Taiwan is JPY 100, what do you think setting up these facilities in Japan would account for?

YKH [00:42:28]: Maybe we have about 35% for that.

ML [00:42:37]: 35% more expensive?

YKH [00:42:38]: Yes.

ML [00:43:02]: I cannot hear you. Could you please repeat that?

YKH [00:43:10]: I think 35% cost efficiencies from Taiwan than Japan because they have opportunities to select their vendors and they already have a resource network. For example, if Fab 1, they don't have enough capacity for some process or some measurement, they can send to other factories next to their fabs, but in Japan, they don't have the resource network built up yet. They can call help from JASM, but I think they might be not that helpful.

ML [00:44:06]: Do you think chip ASPs could be different from those for Taiwan- and Japan-based fabs? Or is ASP the same but cost is different?

YKH [00:44:18]: Cost is different, but in this area, they don't build it in Taiwan, so there's nothing to compare.

[00:44:27] JASM Fab 2 is under construction, and Sony and TSMC are setting up new JV facilities together. Do you think TSMC has enough engineering bandwidth and resource capacity to simultaneously manage both projects?

YKH [00:44:58]: No, because I think they provided 600 positions in Japan for P1, P2, and the they did the AVSM, as you know that they are still expanding in Taiwan and also expanding in America and recently Germany, so they don't have enough engineers.

[00:45:39] You mentioned that workforce shortage is one of the problems. Besides high salaries, what are TSMC and Sony implementing to attract or retain the engineering talent at Kumamoto?

YKH [00:46:05]: Actually, I stay there for the near two years. They only adjust the salary for three or four times because there is a huge problem and they also started to do long-term (inaudible 46.28) from education. They set up the semiconductor academy to work with Kumamoto University and normal and night high schools because even high school students cannot be the engineer, they can be the technicians.

ML [00:46:52]: How many local Taiwanese engineers come to Japan? What percentage do Taiwanese engineers account for in JASM vs local Japanese engineers?

YKH [00:47:13]: Because I joined in early stages, so it's around 30% Taiwanese. Also, the management level, they wanted to decrease the portion of Taiwanese, not to zero, but they still wanted to decrease it. I don't know what is the exact number… Also, the engineer from Taiwan is relatively expensive than that, not only compared to their salary, company will pay the tax for you. Tax also is their cost, not only the compensation given. They also provide the subsidies like housing, like the transportation.

ML [00:48:13]: Maybe the same thing should happen for this new JV, but you are not sure because TSMC is expanding globally.

YKH [00:48:23]: Yes, it definitely will happen.

[00:48:32] Kumamoto faced two big earthquakes. What are the ongoing concerns around that? Have there been any seismic impacts on precision fab operations, or is it the case that you can do factory operations the next day after an earthquake?

YKH [00:49:08]: Actually, I had this situation when I was there (ph 49.10). In TSMC, they have so many experience for earthquakes. They recover and they do a lot of preventions for that. In this kind of event, the event started from Tuesday? I think it's a Tuesday. They recovered on Saturday. The impact is about two or three days capacity. I heard that the next Monday, they already fully recovered and back to normal and the management wanted to check the status of Sony, "What is the status of our neighbours?" Sony told them that, "We are just about to start to do the recovery." Yes. That's a total difference of the culture. At that moment, Sony sent their employees back to home first, but TSMC requested their employees to stay and start the recovery.

ML [00:50:33]: The recovery time is shorter than expected.

YKH [00:50:42]: Yes. I think the estimate recovery for Sony is until the end of August. They spent almost whole month for that.

[00:51:01] How do you assess the impact on local infrastructure? Do you think they can handle TSMC and Sony being in the same place? Besides workforce shortage, are there any shortages in water, electricity, etc?

YKH [00:51:30]: The water and electricity is not impacted… Sorry, water is a little bit impacted. Electricity is fine. The water, I think is impacted for normal human life, but for the industrial level of water, they filter, so that's no huge difference. Not speaking of the earthquake, I heard that the electricity is going to be not enough in Kyushu area, because when I arrived there, they had plans to maybe P3, P4 but the facility team told me that the P3 were constrained by the electricity capacity.

ML [00:52:33]: Thank you very much. We will now conclude this Interview.

YKH [00:52:42]: Thank you.

Transcription ends at 00:52:50 of the recorded material