Joel Monnier: ‘Kalray has decided to reinvent the processor’

Joel Monnier, a former executive of STMicro, is now leading Kalray – a French fabless silicon startup which has been promising chips for the embedded market, based on the VLIW (Very Long Instruction Word) architecture since it was founded in 2008.

Like all embedded technologies, Kalray aims to be very energy efficient. Kalray says it is tens of times more energy efficient than the competition. In this, it is similar to rivals such as Britain’s XMOS. But there are hints that the firm may be aiming for bigger targets: it hopes its multi-core chip could ultimately have applications outside the embedded market, in more general purpose applications.

The dream came a little closer to reality in 2012, when Kalray delivered samples of a 28nm version of a 256-core chip.

Silicon France spoke to Joel Monnier, the head of the company.

Bringing back VLIW

For your MPPA 256, you advertise computing capacity of 230 billion operations per second (0.23 teraflops) with very low power consumption. What is it about massively multi-core technology that helps match or even beat other multi-core processors? Is it related to the adoption of the VLIW architecture?
Multicore processors typically feature four or eight cores. At Kalray, we have adopted a disruptive architecture that enables up to 256 cores in 28 nm process – architecture that, in two years, will allow us to go up to 1024 cores in 20 nm process. Consuming less than 10 Watt, the MPPA 256 can achieve energy savings by a factor of 100 compared to other processor families.

Transmeta, the creator of Crusoe processors, had also adopted the VLIW architecture, but without commercial success. What have you learned from their failure?
The VLIW chips are widely used in the industry, but Transmeta had sought to emulate an X86 processor, a task which generated too many complexities and ultimately led to failure. The MPPA is not in the same market.

How many applications are compatible with your processor? Does your AccessCore developer environment facilitate application porting?
Today, our processors support more than fifty applications, and we actually offer development tools, debugging software and other libraries, so you don’t have any constraints in the development process. Until now, it took nearly a year to create a complex application for an FPGA [Field-Programmable Gate Array]. Working with Kalray reduces development time to just a few weeks.

The grand scheme of things

You are targeting markets underserved by reconfigurable FPGA processors, but which do not justify the development of dedicated ASIC chips. Why not address the market of smartphones, computers or servers that are also seeking the maximum power with minimal energy requirements?
Our priority is to address markets that need small chips with high computing power. We get a lot of interest from manufacturers of signal processing and security products. I do not think our processors will end up in smartphones, but the server market is within reach, partly because of the energy savings that our chips can generate.

The French champions such as Technicolor, Thales, Alcatel, Dassault Systèmes or Bull are likely to be interested in your processors. Are you in contact with any of them?
We are running R&D programmes with some of these companies, but we also receive a lot of interest from United States and Japan. We will announce new partnerships shortly.

Your processors are designed between Grenoble and Paris, but manufactured by TSMC in Taiwan. Can you relocate the production to France?
We seek to create a family of low-cost processors. Working with the world’s leading foundry gives us large volume production capacity, which helps achieve significant economies of scale. But the means of production also exists in STMicro, and relocation is not out of the question.

STMicro seems to have abandoned processors in favour of MEMS (Micro-Electro-Mechanical Systems). Most semiconductor foundries are now located in the US or Asia. What is your model? Are you going to stick to fabless manufacturing? Can it help the European processor industry?
Kalray decided to reinvent the processor through original software architecture and optimization. We are “fabless” but unlike ARM, Kalray sometimes sells hardware and boards. Our model is more like those of Xilink or Altera, and our goal – to achieve one billion euros in sales by the end of the decade!

Euro Story: Each week, we publish a selected story from across NetMediaEurope’s network of European sites. This week’s story by Jérôme Bouteiller is from Silicon.fr. It was translated and localised by Max Smolaks.

What do you know about Europe’s role in Tech history? Take our quiz!