We hear Intel is considering acquiring SiFive for US $ 2 billion (about 220 billion yen). I feel that $ 2B is too small for a first-tear RISC-V company which more or less represents RISC-V itself. SIFive is probably aiming for an IPO on its own terms. If SiFive isn’t profitable for its large number of employees, it may be unavoidable to buy it from Intel and exit. But I feel like Intel acquiring SiFive might not be the best roadmap for RISC-V. In a sense, the value presented for RISC-V seems low.
SiFive is also said to have received acquisition offers from multiple companies, not just Intel. In such a case, SiFive tend to sell out, for example, when the counter-proposal from another company is around at 10B,
ARM became profitable in 1993, eight years after its founding in 1985. Then ARM was listed on the London Stock Exchange and NASDAQ in 1998, the 13th year from its founding. By that time ARM had the monopoly on the mobile phone chip set market. Robin Sachsby, CEO from 1991 to 2001 and Chairman from 2001 to 2006, steered ARM very well.
The instruction set (ISA) business cannot be commercialized until it goes from one end of the semiconductor supply chain to the other. It is the business with very lowest turnover rate.
Jensen Huang of NVIDIA, now the de facto owner of ARM, gave me a valuable opinion when we cooperated in the development of NV2, NV18, etc., ages ago. At the time I worked for a Japanese electronic company in the RISC instruction set (ISA) products. Mr. Huang called this business an instruction set franchise.
According to Mr. Huang, software development is a loss-and-loss operation. If the semiconductor company covers the software cost that supports the instruction set (ISA), then eventually the company will go bankrupt. On hindsight he was referring to Japanese semiconductor companies spending huge amount of money developing Windows CE with Microsoft, At the time Microsoft’s Bill Gates demanded Japanese semiconductor an instruction set (ISA) partners to cover 50% of the development cost to co-created the market. It was a reasonable position then, but in the end, the development cost of Microsoft chose to promote Windows CE with ARM as a sole partner.
ISA franchise should be dominant so most software development is covered by third parties. In some cases it is inevitable for ISA franchise to pay for software at critical points. The RISC-V software development is done by the third-parties and community. SiFive is in a good position.
In the instruction set (ISA) based processor business, there is a hardware design cycle and his software design cycle, which even if you invest more money, the development period often is not be shortened. According to Fred Brooks’ Myths of Man Months, increasing the engineers may reduce development efficiency and even slow down development. In the end, software engineer is the king. Good software engineers can double the processing speed and/or halve the power consumption.
The ARM instruction set (ISA) processor development cycle was one generation ahead in four years. At a domestic instruction set (ISA) maker aiming for one generation per year, the next processor came out before the end of the software development cycle, and I think that software development was rather delayed. Even in the midst of excessive competition, ARM properly aligned its hardware design cycle with the reality of software design cycle.
Recently, Renesas Electronics (Renesas) has entered into a partnership with SiFive for automotive semiconductors, in addition to a partnership with Andes in Taiwan for ASSP products. If this kind of trend continues, I think that both RISC-V and SiFive will eventually excel in PC and server applications in addition to present embedded, IoT and RF applications.
We would still like SiFive to aim for an IPO for the time being.
