TSMC’s Kumamoto Plant: Chip Orders Exceed Production Capacity, Even Before Mass Production

[Daniel Nenni]

TSMC is the world’s largest chipmaker. They make chips for a wide range of industries, from smartphones to cars to data centers. Their chips are in almost every piece of technology we use in our daily lives. This makes TSMC very busy, and they are constantly producing chips. However, even this is not enough, as the production capacity of their new Kumamoto factory is already fully booked before it has even started mass production. Here are the details…

TSMC and Sony Partner to Meet Growing Demand for Automotive Chips
TSMC is always looking for ways to meet the ever-increasing demand for its chips. To do this, the company not only upgrades its existing production facilities, but also builds new ones. One of these new facilities is the Kumamoto factory in Japan. The Kumamoto factory cost TSMC more than $8.6 billion to build. It is scheduled to start mass production in 2024 and will employ around 1,700 people. The project is so big and important that it was a collaboration between TSMC and Sony Semiconductor Solutions.

Sony and TSMC has indicated that the demand for semiconductors will outstrip the supply that can be provided by the TSMC Kumamoto plant. This news has been revealed by Terushi Shimizu, President and CEO of Sony Semiconductor Solutions. The plant has already received a large order from Honda, and the increasing interest in electric vehicles is expected to lead to even more orders. Industry experts believe that the partnership between Sony and TSMC will help the plant reach its full capacity early on.

Moreover, this isn’t the only possible solution. One possibility is that TSMC will build a second wafer production facility in Japan. TSMC’s chairman, Mark Liu, has said that the company is considering this option, but no final decision has been made. If TSMC does decide to build a second plant in Japan, it would likely be located near the current plant in Kumamoto. The Japanese government is eager to attract more investment from TSMC, so it is expected to offer the company substantial subsidies.

https://www.gizmochina.com/2023/06/26/tsmc-kumamoto-plant-production-capacity-fully-booked/

 

[Daniel Nenni]

It seems to me that automotive is a bubble ready to pop?

 

[benb]

This is part of the new trend to build fabs outside of the "big fab" countries of Taiwan and Korea. Japan did not build a new fab for decades, so this is a turning point in the path to rejuvenating the historical (and future) semiconductor industry in Japan.

 

[Tanj]

It seems to me that automotive is a bubble ready to pop?

Why? China is now the largest auto market and is mostly first time buyers or rapidly trading up to better. Many other countries tracking that increasing affluence and mobility. The G20 are rapidly transitioning to EV which will soon include a large transition to self driving, a huge churn for the next decade. From the chip POV, huge growth in new silicon.

 

[Paul2]

The G20 are rapidly transitioning to EV which will soon include a large transition to self driving, a huge churn for the next decade. From the chip POV, huge growth in new silicon.

I don't see
— G20 countries moving to EVs. China, while being decade+ ahead with EVs is nowhere even close to even hybrids being dominant. So we can say fore sure the next decade IC cars will dominate
— EVs require less parts than petrol powered cars, and are often more mechanically, and electronically simpler than modern IC engine powered cars.

 

[count]

Cars are evolving to be basically computers on wheels. There is a trend to do everything by wire.

Car sales are not increasing but semi content in cars is. This is not a bubble but long term trend. That said auto is a very cyclical market.

 

[Paul2]

Cars are evolving to be basically computers on wheels. There is a trend to do everything by wire.

Car sales are not increasing but semi content in cars is. This is not a bubble but long term trend. That said auto is a very cyclical market.

The semiconductor content in cars is decreasing, and the last chip crisis has accelerated the process.

It peaked in around 2010, and been slowly decreasing since. Mercedes of the era easily had 100+ MCUs, and few times that of other ICs. Now it's lower, and the next generation will likely drop the number dramatically.

The music player+amplifier alone had like 25 micros, and a fully functioning risk CPU with MMU in it. Now, low end ones just have a single soc + few auxiliary analog ICs.

 

[KevinK]

@Paul2 , number of chips might be falling but % of car component cost continues to rise, at least through 2030. About 40% of the total BOM cost today growing to 45% in 2030.

How Many Semiconductor Chips Are There in a Car?

I recently read a statement that there is something like 1400 individual semiconductor chips in a typical modern car. I wondered, “Can that be correct?” 1400 is a lot of anything.…

economistwritingeveryday.com

 

[Paul2]

@Paul2 , number of chips might be falling but % of car component cost continues to rise, at least through 2030. About 40% of the total BOM cost today growing to 45% in 2030.

How Many Semiconductor Chips Are There in a Car?

I recently read a statement that there is something like 1400 individual semiconductor chips in a typical modern car. I wondered, “Can that be correct?” 1400 is a lot of anything.…

economistwritingeveryday.com

The BOM cost grew because of microchip industry consolidation, waves of rollups, and the chip crisis

I saying confidently, the "braininess" of an average car is going down, as is the design complexity. That's my first hand knowledge

 

[jms_embedded]

— EVs require less parts than petrol powered cars, and are often more mechanically, and electronically simpler than modern IC engine powered cars.

I would not consider the power electronics and associated electronic control systems required for electric vehicles simpler or less expensive than the electronic systems they replace on a combustion engine vehicle.

 

[Paul2]

I would not consider the power electronics and associated electronic control systems required for electric vehicles simpler or less expensive than the electronic systems they replace on a combustion engine vehicle.

They are, it's if you open Tesla's inverter you will see a ton of FPGAs, and gimmick components, but all newer inverters in EVs see rapid reduction in component counts. It will be in this decade, where we will see it boil down to just a driver ASIC/hybrid, and a few fat switches I bet.

 

[jms_embedded]

They are, it's if you open Tesla's inverter you will see a ton of FPGAs, and gimmick components, but all newer inverters in EVs see rapid reduction in component counts. It will be in this decade, where we will see it boil down to just a driver ASIC/hybrid, and a few fat switches I bet.

I'm not comparing Tesla inverters (with the assumption there of high-end; not sure what you mean by "gimmick components") with newer EV inverters; I'm comparing the inverters and battery charge controllers with combustion engine vehicles that would have an ECM and fuel injection system with pump and valve drivers.

Coolant pump and fan will likely still be there, maybe smaller.

What else in combustion engine vehicles do we get rid of in EVs? (remember, I'm just talking electronics content, not the mechanical systems)

 

[KevinK]

The BOM cost grew because of microchip industry consolidation, waves of rollups, and the chip crisis

I saying confidently, the "braininess" of an average car is going down, as is the design complexity. That's my first hand knowledge

@Paul2 , depends on how you measure “braininess”. If you measure in terms of number of scattered microcontrollers, I might agree. But if you measure in terms of compute power or number of transistors, I disagree. Between increased communication (plus OTA updates) and new safety systems with vision and LIDAR, the numbers all point in one direction - upward.

 

[Paul2]

What else in combustion engine vehicles do we get rid of in EVs? (remember, I'm just talking electronics content, not the mechanical systems)

A modern IC car have few dozen sensors in the powertrain, each has an analog component, maybe something with basic serial i/o, 1-2 MCUs, one which does real time computation, and another doing high level protocol communications over CANBUS, or something else.

Fuel, multiple separate cooling systems, valve timing, lubrication systems each have their own digital components, doing PID on motors, pumps, actuators etc.

A differentialless 2-motor based EV can do without even a hardware ABS, because motor controller knows torque, and RPM on each wheel by default.

 

[Tanj]

Consider also the problem that CANBUS totally lacks security (lookup CANBUS access via headlights) and even if the functionality barely shifts the auto business is facing a wholesale redo of every chip attached to the vehicle bus in the next few years in order to fix security. It is a mess.

 

[jms_embedded]

A differentialless 2-motor based EV can do without even a hardware ABS, because motor controller knows torque, and RPM on each wheel by default.

If you want that kind of behavior, then you need redundancy in the current/position sensors in the traction motor controller, which adds cost. (You can do motor control without position sensors and with a single current sensor, but that's vulnerable to failures, and traction motor is NOT where you want a motor controller failure. Especially if you have a differential-less system where the motors must be rotating at close to the same speed or Bad Things Can Happen.)

Small motors with modest performance/reliability requirements can be controlled with a 6 power MOSFETs + 1 chip (= microcontroller + MOSFET gate drive) + 1 current shunt resistor solution. And a small number of jellybean passives. I work with these kinds of systems for my day job. On the other hand, traction motors in the 50kW+ range, with ASIL-D requirements, are not cheap. I am sure they will improve integration going forward, but the power electronics costs are significant, and the gate drivers for that size motor drive have some serious requirements: dielectric withstand of 1000V for a system with 800V DC link; gate drive output stages have to go up and down from 0 to 800V and conduct on the order of 10-20A for short periods of time from floating gate drive power supplies, in order to turn the power transistors on and off quickly. Unlikely to be something you can put into the same package with a SoC or microcontroller. Definitely not the same die, unless someone knows of some magic semiconductor process that is cost-competitive with something in the 28nm - 65nm range and handles both the very high voltage and the complexity of a high-speed microcontroller with integrated analog.

 

[Paul2]

then you need redundancy in the current/position sensors in the traction motor controller, which adds cost.

Modern motor controllers implement that: 1 hall sensor + current sensing + field oriented control fallback if everything fails. You want an extra sensor? Add 1 hall sensor for 10 cents. Want two extra sensors, add 2 extra hall sensors. Want an overkill? Add an inductosyn.

Definitely not the same die, unless someone knows of some magic semiconductor process that is cost-competitive with something in the 28nm - 65nm range and handles both the very high voltage and the complexity of a high-speed microcontroller with integrated analog.

https://www.hitachi-power-semiconductor-device.co.jp/en/products/ic/inverter_ic/index.html very far from logic IC node though, and only 1A, but powerSoC are definitely a thing. Even megawatt scale switching is getting integrated, Panasonic is working on that now.