Blog

Jun 06, 2023

3nm AI chips and 6nm microcontrollers will be key to TSMC Dresd...

TSMC’s European Technology Symposium was heavily focussed on automotive, and of

TSMC's European Technology Symposium was heavily focussed on automotive, and of course the prospects for a fab in Dresden.

The company says it is carrying out due diligence on the decision, which is building the business case. This will go to the TSMC board, with the board meeting in August as the earliest opportunity.

"This is still ongoing," said Kevin Zhang, senior vice president for business development at TSMC. "To be clear we are going through internal due diligence right now and there has been a lot of good progress with support from local government and EU government."

But there are three key trends that impact on that business case.

The first is TSMC's aggressive move into 3nm FINFET production for automotive AI chips. An automotive version of a process usually takes two to three years until chip designers can use the process. TSMC has now launched the development kit (PDK) N3AE process, or ‘automotive early’. This combines the artefacts and the ’best guess’ for designers to get started now.

This is one compelling argument for Dresden supporting a 3nm process. With AI chips for advanced ADAS and self driving cars and robotaxis is series production in cars made in Europe in 2025, avoiding a repeat of the pandemic shortages for European car makers is one of the primary goals of the fab. And that supply chain will include 3nm AI chips from a number of suppliers that would otherwise be made in Taiwan or in Arizona.

"We think Europe is very significant, that's really the main factor why we want to build a fab in Europe as well as the talent resources," said Zhang.

But there is another key technology trend.

"The goal is to be close our customers and in Europe half our business is microcontrollers," said Zhang. "If we do build a fab in Dresden it is likely to be the 20nm generation with embedded RRAM," he said.

However the technology symposium showed plans for a dramatic jump for microcontrollers. The company is developing a 6nm process for microcontrollers. This would be a dramatic leap forward from 28nm today would be a critical process for the Dresden fab.

"N6 is further out than 2026," said Zhang. "MCUs are only just moving to 16nm and usually it takes quite a few years to ramp up. In 28nm its probably five year and then it will be 6nm."

This means the Dresden fab will need to support 6nm in 2027. If a decision is made by the end of the year, which would be fast, then building the fab, ordering the equipment and qualification would be 2026.

These will be compelling arguments for support from the EU CHIPS Act.

The third is TSMC's position in the industry. At the Symposium, CEO CC Wei highlighted that TSMC does not compete with its customers, pointing out the difference with his competitors. Although he did not name them, this is Samsung and Intel Foundry Service for the advanced technologies, rather than the specialist suppliers such as GlobalFoundries and UMC.

TSMC's fab building are not process dependent, it's about ordering the equipment from ASML and other suppliers.

The company usually wants to own 100% of its fabs, but it is open to collaboration in Europe, says Zhang, pointing to the joint venture fab in Japan with Sony and Denso currently under construction. Although he would not comment on potential partners, these include Bosch, NXP and Infineon, as well as long term partner STMicroelectronics, who are all key for automotive microcontrollers and would be looking at the N6 6nm process. ST's CEO Jean-Marc Chery and NXP's CEO Kurt Sievers met with Wei at the symposium.

By the time TSMC Dresden comes online the 3nm process will be well established with higher yields and lower costs. The leading edge will be the 2nm nanosheet gate all around technology that will be in production in Taiwan and work will be advancing on the production version of 1nm technology

www.tsmc.com