“After I left Arm, many of my former colleagues at Arm went to various companies working on RISC-V, and they all told me that RISC-V is very interesting.”
Arm was also very small back in the day.
“Arm is a great company,” Wu Xiongang said without hesitation. “But just like RISC-V, Arm was once very small. For most of the time, the CPU market was dominated by the ‘Wintel’ alliance.” Wu Xiongang continued. However, after the iPhone was released in 2007 and the Android system beta was unveiled in 2008, Arm’s historical trajectory was rewritten.
Arm’s past advantages are now becoming its disadvantages.
“Arm’s standardization was designed to support numerous mobile applications, requiring all users to adhere to specifications, which does not allow for faster innovators like Wei-Han to break through. Since their main revenue comes from existing large customers, Arm must also ensure compatibility. In other words, it’s not that Arm is unwilling, but rather that it cannot, which is the core issue,” Wu Xiongang explained.
He stated: “Arm is suitable for standardized products. For example, Arm’s engines come in four-cylinder, six-cylinder, and eight-cylinder variants, but they are all designed for the old track. On the new track, you can only use these and cannot modify them. However, in the new field of AI computing, where ‘racing’ is the focus, everyone’s track is different.”
In the face of the surging wave of AI, Wu Xiongang commented: “Just like the mobile revolution, AI has new user groups that will generate new data: the PC era produced work data, the mobile era produced photo, WeChat, WhatsApp, and other information data, while AI and autonomous driving generate camera data and real-world data, with the data users shifting from ‘humans’ to ‘machines.’”
“Different data leads to different computing demands, thus necessitating new architectures,” Wu Xiongang continued. According to his view, in the next 5 to 10 years, entirely new architectures will inevitably emerge. Just as the personal PC era could not continue using the IBM Power architecture, and the mobile era could not continue using Intel’s x86, the AI field will also struggle to use x86, IBM, or Arm architectures. The reasons are as stated above—data and application scenarios have changed, and the problems that need to be solved are different, making it impossible to use the same computing architecture.
“Arm has been very successful and large, and it remains a great company,” Wu Xiongang reiterated. However, as mentioned, AI represents a new and larger technological revolution that requires new architectures, and RISC-V is becoming the optimal choice for this track.
What is he doing?
Wu Xiongang has chosen to start with custom processors, and the startup CoreLab, where he serves as chairman, is the main vehicle for this new journey. According to its official website, CoreLab is a company focused on providing custom processor IP solutions that not only possess Arm-level quality but also the flexibility of RISC-V. CoreLab aims to solve the critical last-mile issues in the IP productization process through comprehensive services, ensuring that products successfully enter the market and achieve success.
“Our mission is to promote open-source-based IP solutions and support our partners’ success through detailed end-to-end support. The services we provide include licensing, CPU performance and area optimization, rigorous verification processes, development tools, and custom instruction extensions,” CoreLab stated on its website.
“SoC iteration is an ecosystem engineering task, and we cannot do everything. The key is the licensing model. We hope to be more open and collaborate with more people, allowing customers to easily obtain the optimal solutions—rather than having customers use three different cores on their chips, with interfaces and tools that are not compatible,” Wu Xiongang said.
Collaboration with Tenstorrent is a wise choice for Wu Xiongang’s new venture.
In Wu Xiongang’s view, designing a CPU is as challenging as building a rocket, and not everyone can succeed. CPUs, like water and electricity, cannot be easily constrained by others, especially when it comes to high-performance chips that can be seen and touched. This is precisely the product that CoreLab and Tenstorrent are working together to create.
“Arm was also not mature back in the day. When I first came to China in 2006, Arm chips could not even run a browser, let alone go online. It wasn’t until Apple released the first-generation iPhone that Arm chips could go online. The subsequent development of Arm is well-known to everyone,” Wu Xiongang recalled.
With this luxurious lineup driving the momentum, let us look forward to the arrival of the turning point for RISC-V in five years.
—- This article is excerpted from an interview with Banhangguan