We attended the RISC V Summit today. We plan to do a deeper dive into how that ecosystem is doing broadly, but here we want to focus on one specific area of interest – the use of RISC V in smartphones.
By way of background, RISC V is an alternative instruction set architecture (ISA), which we have covered extensively. One of the big challenges facing RISC V is that existing products have decades of software built for other ISAs. In mobile, that ISA is Arm, Switching from Arm to RISC V would be very difficult because so much of mobile software is written for Arm that the effort of switching would be immense. Here software is defined broadly encompassing operating systems (OS) and applications, but also the low level firmware that chip designers provide as drivers interfacing between the OS and the processor, as well as the tools those chip designers use to design, build and test their chips. As a result, we tend to be very cautious around claims that RISC V will really take off in mobile.
That being said, it looks fairly likely that some time soon we will see mobile application processors built on RISC V in smartphones.
For starters, Google has been very vocal about promoting the use of RISC V for Android. Their motivations for this are not entirely clear, partially the appeal of “open” and partially wariness around its reliance on Qualcomm, and probably a few more factors. If we had to guess we suspect that they will launch a version of their Pixel phone using an internally designed applications processor based on RISC V in the coming years.
At the RISC V Summit, we attended a string of presentations on this subject of RISC V in mobile. In many senses, these talks did not inspire much confidence in the effort. A speaker from Samsung spent his entire time describing the very labor-intensive work involved just to get a benchmark test working in a way that was compatible with RISC V. Again, all that work just to get the testing tool to work. By the third mention of “hand coding” we were convinced that he was going to say they had abandoned the effort. However, at the end he said that they finally got their processor working and seemed to imply that it would be available in “non-flagship” models soon. Then a speaker from SiFive got up. They have been working hard on Android, and he started his presentation by stating that Android on RISC V was absolutely coming, but then he went through his own litany of woe. He detailed the many missing software drivers which SiFive then had to write themselves, almost of all which he said ‘would not be merged’ into the main Android code body. It got so bad that we grabbed the only question spot to make sure we had heard him correctly when he had said “Android is coming soon”, which he confirmed. Finally a speaker from Alibaba’s chip division T Head took his turn at crushing dreams with a lengthy discussion of the intricacies of the Android software environment. Again, we were left uninspired, except in the T Head booth they have a working demo of Android running on their TH1520 (see photo above).
In the end, we read this cavalcade of misery as a form of signaling what these companies need Google to do next. In fairness, there is a chicken and egg problem at work here. The hardware people need fixed timelines for Android API features. At the same time, Google needs to get clarity from the hardware ecosystem on their timelines.
And then there is Qualcomm. They have already launched a chip for Android wearables, and have ample motivation to stick a finger in Arm’s eye. We missed the keynote from Qualcomm’s RISC V point person Ziad Ahsgar, but he did not seem to make any headlines. Their Arm software library is the largest so the cost of them switching is highest. Nonetheless, we have to think they are thinking very hard about a switch like this.
So Android on RISC V works, Alibaba proved that. Samsung already has transceivers running RISC V and have an application processor ‘close’ to production. Despite all the challenges and many lacking features, it sounds very likely that we will start to see these systems in the wild soon, Maybe in 2024, maybe in 2025, and definitely with a whole host of growing pains.