Skip to content
News & Analysis

Chiplets: If It Happens in China, Will It Stay in China? 

Going beyond trade barriers, what's driving China to chiplets is financial motivation. They are looking for the reuse of low-cost, low-risk chiplet IP.
Chiplets: If It Happens in China, Will It Stay in China?

Share This Post:

By Junko Yoshida

What’s at stake:
China’s chiplet aspirations are well known. But will China use its domestic chiplet activities as an opportunity to decouple from the rest of the world? We think the reality is contrary.

Chiplets represent a once-in-a-lifetime revolution that will allow the semiconductor world to disaggregate the processes of designing, manufacturing, testing and packaging silicon – the fundamental matter who does what. China wants to seize this moment to play a big role in this new world order.

To many Western observers, however, it remains unclear how far Chinese semiconductor vendors have already gone with domestic chiplets development. Also unknown is  the applications for which China plans to leverage chiplets. Nor is anyone certain whether China’s chiplet embrace reflects its desire to break free from U.S. trade restrictions that have focused increasingly on technologies.

The Ojo-Yoshida Report recently sat down virtually with Wayne Dai, CEO of VeriSilicon, for the upcoming episode of our Dig Deeper on Chiplets podcast series.

VeriSilicon, based in Shanghai, has established itself in semiconductor IP licensing and custom silicon design. It is well connected not just with domestic customers but also chip designers worldwide.

Our discussion with the CEO reveals that China’s approach to chiplets is neither parochial nor narrowly focused as a recourse for export control.

China is getting ready to play in the global chiplet market.

VeriSilicon is a case in point. Recognizing significant interest in chiplets from the auto industry, it is bringing together several carmakers, including a top European brand, to tailor link-layer specifications for chiplets in automotive applications.

We should have international standards for chiplets. VeriSilicon, for example, is one of the first to join UCIe.

Wayne Dai

This move is significant for several reasons.

First, China is not developing its own chiplet standards for the sake of being different. Dai told us, “We should have international standards.” VeriSilicon, for example, one of the first to join UCIe, has been active in its work.   

Second, Dai recognizes the importance of developing “domain-specific” link-layer protocols for chiplets. While the physical layer describes the way data is transmitted between chiplets, the data link layer is needed on top of PHY to define the protocols.

A chiplet is not easy. Once defined, it cannot be changed. “Unlike software IP – deployed in SoCs – chiplet is a hard IP,” explained Dai.

Defining chiplets requires domain-specific knowledge. But once engineers know that  a certain chiplet is going to be used for automotive, for example, it gets easier for them to define specs for automotive chiplets.

“So for automotive, we should have a link layer on top, optimized just for automotive. I think that’s fine,” said Dai. After all, there’s no way that we can make a one-size-fits-all chiplet, he added. “Otherwise, overhead becomes too big.”

Five companies working to define automotive-specific chiplet link-layer protocols include a top European automaker. These companies are sharing the cost upfront.

Wayne Dai

Third, VeriSilicon is fully aware that neither purely IP companies nor purely design service companies can do chiplets. Chiplets must be defined by customers who know the domain.

As Dai explained, five companies working to define automotive-specific chiplet link-layer protocols include a top European automaker.

Last but not least, VeriSilicon is taking a practical business approach to pull things together. Time, effort, and resources required to develop link-layer protocols will be shared among the five companies. “If they know they have to pay upfront, they do get serious,” said Dai.

But once protocol specifications are complete, the group will make them available to other auto players willing to pay royalties for chiplets.

Chiplets are fundamentally different from SoCs, in the sense that not every company working on chiplet development needs a 5nm tape-out. Once the non-recurring engineering cost of a chiplet is covered, the same chiplet can be reused by others.

China’s native chiplet interface standard
In 2022, China unveiled its first native chiplet interface standard, called “Technical Requirements for Chiplet Interface Bus.” This standard, officially passed the review of the China Electronics Standardization Association, has been released by the Ministry of Industry and Information Technology in China.

According to VeriSilicon, the architecture of the China’s indigeneous interface standard includes the Data Link Layer (DLL), Physical Adaptation Layer (PAL), and Physical Layer (PHY).

More specifically, the standard supports both serial and parallel physical layers.  

The requirements for power consumption and delay, described in this Chinese standard, are very low, making it convenient to adopting right away, added VeriSilicon.

China’s chiplet interface standard remains compatible with UCIe on the PHY level.

Triggering China’s excitement over chiplets, noted Dai, revolves around the potential of chiplets to drive down the cost of chip design.  

During our discussion, Dai also mentioned that China’s chiplet interface standards, aside from supporting both for 2.5D and 3D advanced packages, are designed for conventional packages.

Imec automotive chiplet activities
As Bart Plackle, Imec’s vice president for automotive, told us in a separate interview, chiplets are seen as the only hope for the automotive industry to get the massive computing performance it needs into cars.

While explaining Imec’s collaborations with key automotive players on chiplets, Plackle cautioned that UCIe alone won’t solve automakers’ chiplet development needs. The interconnect standard should “allow variations in protocols, reliability measures, security, and other layers vital to the automotive market,” he pointed out.

Last fall, Plackle counted more than 50 companies at Imec in Leuven to hash out their problems.

Obviously, with so many companies involved, “it’s not easy to streamline and get to a standard,” he acknowledged. The gathering ended up identifying four priorities, with a working group for each: architecture, security, protocol and reliability. Plackle noted, “Those working groups need to clear their thoughts, and decide what needs to become part of the standards.”

It’s not clear now how the auto group mentioned by VeriSilicon’s Dai relates to the Imec-led initiative.

The podcast with VeriSilion CEO Wayne Dai is scheduled for broadcast at 9:00AM PT, on March 27, 2024. Register here to save your seat.

Bottom line:
The chiplet is a nascent market. China is determined to be a player. It’s important to note that the real value that China is seeing in chiplets are the eventual “reuse of  low-cost, low risk IP,” according to Wayne Dai, CEO of VeriSilicon. One thing is for sure. What China decides to do won’t stay just in China.

Related stories:


Junko Yoshida is the editor in chief of The Ojo-Yoshida Report. She can be reached at junko@ojoyoshidareport.com.

Copyright permission/reprint service of a full Ojo-Yoshida Report story is available for promotional use on your website, marketing materials and social media promotions. Please send us an email at talktous@ojoyoshidareport.com for details.

Share This Post: