Thoughts after COMPUTEX
Written by: Jack Lau
We just came back from the COMPUTEX show in Taipei last week. COMPUTEX is one of the best computer hardware shows.
Other than exhibitions by the like of INTEL, ASUS, MSI, and other big names, there are a number of panel discussions. One of the panel discussions is about a processor called RISC V.
Before we explain more and why we should pay attention, let me digress and talk a bit more about a Cambridge based company called ARM.
The company was founded 28 years ago in England. Its last reported revenue was JPY152 billion (about US$1.4 Billion).
If one has followed the recent news on the trade dispute, the name ARM probably had popped up a few times. https://www.wired.com/story/huawei-loses-arm-chip-design/
Most noticeably, it would be the suspension of support of ARM to Huawei. So, let’s take a deeper look.
Semiconductors these days are rather complicated. To ensure success in design, one probably needs a few things: (a) an excellent Electronics Design Tool (“EDA”), (b) a good design library, and (c ) an excellent foundry (the factory).
A good design library is like a “function” in an Excel spreadsheet. In an Excel spreadsheet, after entering rows or columns of data, you can elect to perform computation, anything from average (statistics), plots, to interest rate, etc. Although all the computation, technically can be computed with basic arithmetic (addition and subtraction), we expect our computation to be done with just a click of a button — or simply select the right “functions”.
In semiconductor chip design, that is called a design library. For instance, if you want your chip to include a USB port or a small memory block, you would like to pick something from the design library and just dropinto the design.
The design library usually has been tested before and it is also specific to different technologies that are related to the foundry.
In case you would like to design something with a processor in the chip, you will then pick a good processor from the library and drop in your design.
This is where ARM comes in. It specializes in processor block. There are a myriad of blocks. To be useful, usually you want to look for something that is low in power, compact in design, and versatile in functions. All these must be tested again and again before you put in a design.
A design run (the process of submitting a design to the foundry and then getting the chip back) is not only costly, but takes months. So, the library must be very reliable.
ARM is indisputably the leader in the processor library sector. It probably owns nearly 90% or more of the market.
So, when we see a company rolls out a chip that has a somehow more sophisticated chip, such as an AI chip for smart phone, it has ARM processors in it.
What are the alternatives?
Here is where RISC V comes into play. To start something from scratch — to build your own processor library is very difficult if not impossible. Imagine the years of experiments and testing before one can safely assume a library part is useful.
One viable alternative is to rely on a processor that is free to use (Open Source). RISC V (“Reduced Instruction Set Computer”) version 5, is a project started at the University of California, Berkeley in 2010. Berkeley of course has a long history of computer chip and operating system design.
And, not to frighten anyone, just having a hardware processor core is not enough either. The processor must come with verified software and be compatible with EDA tools and foundry services.
RISC V has begun to gain some traction. https://www.eetimes.com/author.asp?section_id=36&doc_id=1334306#
Even the like of Nvidia is reported to adopt RISC V in some of their chips. https://www.youtube.com/watch?v=gg1lISJfJI0
*Caveat: While processor core from Berkeley is “free” and Open Source, as in the case of some Open Source technology, there are some private companies which make the Open Source more commercially viable (better support). So, to use the RISC V core, one may still have to “license” from 3rd party company. But the fee is significantly lower. While an ARM core licensing fee can be up to US$3M per core, that of a RISC V core is probably about US$250K per core.
In this edition, we want to start this experiment by listing (partially) the companies that may be related to the current blog. Let us know what you think.
Huawei, ARM, Synopsys, TSMC, Softbank