The European Processor Initiative

At the recent RISC-V Global Forum, one of the presenters was Jean-Marc Denis, chairman of the European Processor Initiative (EPI), or as he described it, the European approach to the exascale age and the road towards sovereignty.

The project is driven by the European Commission. Jean-Claude Junckers, in 2015, then the president, said that:

Our ambition is to become one of the top 3 leaders in high-performance computing by 2020.

I'm not quite sure how you'd go about measuring that objectively, since the #1 supercomputer just announced in this year's top 500 list is Japanese. Most of the rest of the top 10 are US and Chinese, although there are three European systems (two in Italy and one in Switzerland). So close, at least. But Europe is so big that if you look at Europe, US, China...then you just have India and Australia to beat!

The EC now has new ambitions to enhance Europe's technological security:

• Blockchain, HPC, algorithms, data-sharing and data-usage tools
• 5G networks
• AI and digital services act
• Single market for cybersecurity
• Joint cyber unit
• Digital education action plan

The EPI is part of this effort. I have to admit that until the RISC-V forum I'd never heard of this, despite having grown up and been educated in Europe, and worked in France for over five years.

In particular, to create an EU-designed microprocessor to empower the EU Exascale machines (in the medium turn). But in the longer term, Europe feels it needs sovereign access (not at risk of limitation or embargo by non-EU countries) to high-performance, low-power microprocessors from IP to semiconductor products. EPI was set up to fulfill this objective.

In the short term, EPI will be Arm-based. Depending on how you squint, Arm is European (except that the UK is withdrawing from the EU), or Japanese (Softbank), or American after the recent announcement of NVIDIA's acquisition of Arm. But the mid-term objective is to use RISC-V. You may or may not know that the RISC-V Foundation moved to Switzerland. In the meantime, EPI has set up a French-based company called SiPearl as the commercial arm of EPI. So the short-term goal is general availability of the Arm-based Zeus core in 2022, and then using that to construct EuropeanHPC exascale machines.

The current state is that there are 27 partners from 10 countries involved in EPI. SiPearl is the industrial and business side of EPI. It is based in France and is fabless. It licenses IP from the partners and develops its own IP around it, and licenses missing IP from the market. They are raising (have raised?) €100M. They will market the chip, and generate revenue, and then work on the next generations. Jean-Marc said that the goal is for SiPearl to become the "European Intel".

The top-level system architecture is in the picture above. Confusingly, the accelerator is called Titan, which is also the name of Google's open-sourced security processor (and a very old computer built at Cambridge back in the early days of computing).

By 2022-23, the goal is to have a European general-purpose processor for HPC machines developed by a European company (SiPearl). The fundamental IP for a European supercomputer based on RISC-V can be developed within the EU. There will be expertise for developing high-end and complex processing units after decades of disinvestment, ready to move onto the next stage of general-purpose processors.

But there are huge challenges, too. To my mind, one is money. €100M is not much in the context of building a state-of-the-art processor from a standing start. it will take five or more years. Some aspects are very difficult to develop and the people who have the capability are rare and expensive. 

Some other challenges Jean-Marc called out:

• Nobody knows where the processor market will be in five years
• Competition (x86 and Arm) comes with a mature complex ecosystem and is well-established
• Between now and 2025, other RISC-V solutions might be brought to market and fail, scaring off investors
• RISC-V ecosystem could become fragmented like Linux
• Private investment on the scale required is rare in Europe

Personally, I'm skeptical about these kinds of projects with too many companies, and too many countries, and too many politicians. Arguably AMD is the second Intel (although not European) but it's taken it decades. Arm has been trying to get into the data center for a decade, too. The most ambitious RISC-V project I know of is Esperanto, but I'm not sure what's happened to them, they have been very quiet and didn't present at the forum (and I see Dave Ditzel is no longer CEO, although he is executive chairman so that might not be significant). They have also been going for five years, longer than SiPearl plans to take to deliver chips.

I'm also a big believer in the way to move technology around being inside people's heads. So you need to get a concentration of resources in one place. There is no one "Silicon Valley" in Europe. There is lots of telecom in Sweden and Finland. A lot of technology around Cambridge (and Arm). Sophia Antipolis in the south of France. Munich in southern Germany. Eindhoven in the Netherlands (Philips). Dresden (GLOBALFOUNDRIES, I think the most advanced fab in Europe although only 22nm). Of course, there's not really just one Silicon Valley in the US anymore. Boston and Austin/Dallas also have critical mass, and Intel does a lot of development in Oregon (and Israel). IBM in upstate New York. But there are thousands of engineers at each of these locations.

It seems unlikely that this will be the "European Intel" in just three to five years. With all the delays in its roadmap, even Intel is struggling to be the Intel of the processor world at the moment. It would take total focus, a team whose reputation and financial reward comes from making it successful. I looked up Jean-Marc's LinkedIn profile. He spent 16 years at Bull, France's national champion in mainframes (and a big VLSI Technology customer, as it happens) and now is head of strategy at Atos. I probed a bit more. Atos acquired Bull.

So let's see in 5 or 10 years if there is a successful European Intel.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.