Stepping into the Indian Institute of Technology, Madras environs feels like you are in a walled jungle. The landscape is full of thick trees that have set down roots for decades. Bats, deers, and monkeys, seemingly comfortable with the few thousand bicycle-riding students, academics, and staff on campus, are easy to spot. The air, thanks to a recent bout of rains in Chennai, is as good as a hill station — very unusual for the port city.
I’m here to meet the team behind the Shakti Processor Project, an ambitious effort in its computer sciences department to build six variants of processors to power Internet of things (IoT) products to high-end servers. If successful — physical manufacturing of chips is to start in the first quarter of 2018 — it will be the first open source processor of its kind built from scratch out of India.
And, if it finds takers in the market, it could make changes in the imports-dependent electronics ecosystem in India, which is predicted to spend more on importing electronics than on oil by 2020.
The Shakti project is based on the RISC-V ISA started at UC Berkeley. For those unfamiliar with the term, RISC, which stands for reduced instruction set computer, is an architecture that uses fewer computing cycles per instruction making it ideal to carry out a set of smaller and general instructions. ISA is short for instruction set architecture, which defines the specifications to be met by a processor.
A C-class controller, an entry-level processor with use in IoT, smart cards and security applications, will be the first in the Shakti series out first quarter of 2018
The first chip of the Shakti series will be a C-class controller chip, an entry-level processor, which would find use-cases in IoT, smart cards, and security applications.
To be sure, there are similar efforts on at C-DAC, or Centre for Development of Advanced Computing, India, under the Ministry of Communications and Information Technology, which is also funding the Shakti project. The C-DAC initiative received $45 million funding, according to news reports in early 2016.
“We started this effort four to five years ago. We have now evolved a roadmap for it — we will go from a controller class to a server class,” says V Kamakoti, who along with G S Madhusudhan, is a coordinator of the Shakti project. Kamakoti is a professor in IIT Madras’s computer sciences department and Madhusudhan is a senior project adviser. Both are members of India’s AI task force,
There are purely academic reasons for Shakti processors to exist. The department’s processor architecture research, for instance, was dependent on scraps of projects it got from businesses and industry. “We did not have a commercially equivalent platform, an effective public platform on which to do research,” says Kamakoti. “For example, if I have a new branch predictor (a unit inside a CPU) and I don’t have an Intel architecture, where can I fit my branch prediction and say, ‘Yeah, it’s working great’? I did not have access to the design, to compare it, with an equivalent in a commercial processor.”
There are also security reasons for wanting to make a processor for India. “We don’t know really whether the processor we are getting from outside is trustworthy. Is it secure?” asks Kamakoti. “Suppose I want variants of a processor, for different needs – not just strategic, even civilian needs – I have to basically rely on the processor available to me, and fit my application to that. It is something like I bought the slipper and I am cutting my feet to fit into it.”
At the Laboratory of VLSI Design, the team is working on the C class of Shakti, which is suited for microcontrollers. Part of what enables a small team to do so is that the processor is based on the RISC-V architecture, an open standard. The entire software toolchain effort is borne by a few universities — primarily Berkeley, Cambridge, and MIT.
Neel Gala, a PhD Scholar from IIT Madras, an early member of the Shakti project told us why a small team, that numbers in the tens, can ship a processor, which would otherwise take hundreds of people.
The RISC-V ISA got traction when the Berkeley team released a processor (in July 2016), which was taped out and running Linux, Gala says. “This was a team of seven students. A team of just seven people working on a new language, on a new processor, was able to beat ARM or its equivalent. That gave us an extra boost, that we can (do it) as well. There is nothing that is sacred and holy that we don’t know.” ARM is a British chipmaker with a dominant share in processors used in mobile phones and tablets.
The team uses Bluespec System Verilog, a hardware description programming language used in chip design, which helps in a much shorter product release cycle. “Bluespec gives you a high-level abstraction, like going from assembly (level programming) to C. You don’t do the dirty work, the compiler does all the work for you. You work at a much higher level, your throughput increases. Simulation and turnaround time increases, your product release time has a shorter cycle,” Gala says.
The RISC-V standard offers quality, flexibility and low cost, setting it apart from proprietary ISAs. The RISC-V Foundation, a non-profit corporation, which wants to make the ‘Linux of microprocessor architectures’, has been getting traction lately. Started in 2014, its membership base has grown to over a hundred organisations, the foundation said in an announcement earlier this week. Apart from the world’s leading tech universities, its members include Google, Huawei, Micron, Microsemi, NVIDIA, NXP, Qualcomm, Samsung, SiFive, among others.
Referring to a Berkley project where a team of seven researchers taped out a chip equivalent to an ARM processor, PhD scholar Neel Gala says, “There is nothing that is sacred and holy that we don’t know.”
There are enthusiastic buglers in this movement of sorts in the microprocessor world. “The capitalist computing bourgeoisie want to enslave us all with proprietary processing architectures, but the proletariat eventually produces its own processor alternative – an ISA for and by the people, where instruction sets aren’t subject to the whim of the royalty-driven class, and where licensing fees don’t oppress the workers’ BOMs (bill of materials),” writes Kevin Morris in the Electronics Engineering journal, lending colour and gravitas to what’s at stake in the processor industry.
Some of that open source zeal can be seen in the Shakti team here. Gala, who was offered a ‘good seven digit package’ by one of the big chip giants, decided against it. “This was an interview over a cup of coffee. The moment of realisation for me was sitting in the cafeteria, and seeing a hundred other PhDs there. The only distinction I had over them was Shakti. If I left it, I would just be another ball in the bag. So that’s why I didn’t leave,” he says.
Shakti has its talent locked in, clearly, but the big imponderable in the chips business is time. Delays are common, even more in academic efforts. So too with Shakti project. An EETimes story in January 2016 said tape out would happen that summer but as our reporting shows, it is now set for just before summer of 2018.
India has been developing processors for nearly 20 years, what’s new is that Shakti is the first effort to consolidate that, so that it is systematic, says Madhusudhan, the second coordinator of the project. “A homegrown processor family, as a consistent effort to spread homegrown processors across the country — this is the first organized broad effort,” he says.
The project could enable the production of homegrown CPUs that are open, patent-free and royalty free. If that sounds like heading into a commodity market, Madhusudhan begs to differ: “If you do commodity processors, it is a commodity industry. If you do state of the art secure processors, AI and ML processors, it is not a commodity industry… high-security processors is not,” he says.
“We can get very customised chips done at far cheaper cost than anybody else. Just like Tirupur does it for knitting, we’re doing it for processors. We’re also trying to incubate companies for handling specific portions of the task.” says Madhusudhan, adding that his students might work on a couple of startups of their own.
He shares an example of how the government is pushing for energy efficient fans and fridges, an area where the Shakti RISC-V processor could find a market-fit. “These use a brushless DC motor, which requires a small processor. We’re talking to one entity that does close to a million fans a month.”
“We can get very customised chips done at far cheaper cost than anybody else. Just like Tirupur does it for knitting, we’re doing it for processors. We’re also trying to incubate companies…,” says G S Madhusudhan of IIT, Madras
An industry veteran says the need for a domestic chipset is staring India in its face. “Our consumption is so mismatched with our indigenous development which means that we are constantly dependant on somebody else to supply, for all our needs,” says P K Sundarajan, formerly the director of IP with Cosmic Circuits, a Bengaluru semiconductor company that was snapped up in 2013 by US-based Cadence Design Systems, a provider of tools and solutions for the semiconductor industry.
The know-how to do these things definitely exists in India. It’s just a question of putting the minds together and getting it done, Sundarajan says. “If you look at some of the Snapdragon cores, maybe its variants, I’m fairly certain it was developed here in Bengaluru. If I look at whether raw talent is available here, there is no question at all. Even with Intel, a lot of the server processing cores probably happens here and likewise with AMD,” he says.
Prakash Mohapatra, an industry analyst, says that he has seen a lot of traction for the RISC-V standard in the last 5-6 months, which has led to a change in his earlier belief that it will not be able to dent the processor ecosystem.
He also points out that over the last few years, ARM has shown interest in enabling custom SoCs at a low investment, through its DesignStart program. (SOC is short for system on a chip, which is an integrated circuit with multiple computing components on a single substrate and is designed for specific applications. They are relatively cheap and consume less power.)
A slowing down of Moore’s law will drive SoC vendors to adopt RISC-V ISA, says Mohapatra. “As innovation in terms of processor technology is not happening, many companies are thinking of how exactly they can improve their performance without compromising on power. All this customisation can be a way of enhancing process nodes. Instead of newer processors, which are really expensive, you can go for custom SoCs, which are tuned to your application,” he says.
While ARM dominates the smartphone market and Intel dominates the server market, a dominant player has yet to emerge in the areas of IoT and AI, areas where RISC-V could make inroads if an extensive ecosystem is in place, Mohapatra says.
“My view is that it will definitely go out of this academic thing and it will go towards a commercial distribution and can actually pose a major threat to ARM,” he says, cautioning that Shakti will have to climb a steep learning curve that is decades old.