Shafy Eltoukhy is a seasoned engineer, with over 35 years of experience in the semiconductor world. He began his career shortly after graduating with a PhD in Device Physics from the University of Waterloo in 1982 to work with Intel Corp as a senior device engineer. Shortly after, he parted ways in order to co-found Actel, an FPGA company now known as Microsemi.
Since then, Eltoukhy has been involved with establishing other silicon and semiconductor companies, including Lightspeed Logic and Open-Silicon Corp, both of which specialize in ASIC design and production. In 2014, he returned to Microsemi as Vice President of the Analog Mixed Signal unit before eventually joining the SiFive team, where he now leads the DesignShare program—an initiative that intends to bring down the barriers to innovation in System-on-a-Chip (SoC) design.
Dr. Shafy Eltoukhy, Head of SiFive's DesignShare Program. Image courtesy of Dr. Shafy Eltoukhy.
“By training, I’m a start-up guy—this is my fourth startup company”, Eltoukhy told AAC during the interview.
AAC writer, Chantelle Dubois, had a chance to speak one-on-one with Eltoukhy to learn more about the DesignShare program, the benefits to IP sharing, and what he believes will be important for the future of SoC design.
Chantelle Dubois (AAC): How did you get involved with SiFive?
Shafy Eltoukhy (SE): The CEO of SiFive called me and told me that he had a great opportunity. I was very excited to get involved with SiFive because I had been in the ASIC business for a while and I know about all the pains that ASIC customers and designers go through, the time it takes to get products to markets, and so on. I saw SiFive as an opportunity to fill in the gaps and resolve the issues that designers and customers have been facing over the last 25 years.
AAC: How did the concept of the DesignShare Program come about and what problem is it addressing when it comes to SoC design?
SE: If you look at any custom design, especially if you are building an SoC, there are few components that you need to have before you start actually building the chip. Of course, you may have your own secret sauce, or your own recipe, or your own IP that you are trying to build, but you can not build a chip by itself based on your idea alone. You really need to be able to use third-party IPs with your own IP so that you can differentiate yourself—a large portion of the costs of building an SoC are the third party IPs.
For example, you will need a processor, you will need I/Os, high-speed interfaces for DRAM, and so on. By the time you add the costs of all the IPs up, you may end up with a few million dollars just to license IPs from third parties. That becomes a barrier for a lot of good ideas from startup companies or from two guys in a garage who want to build their own idea but can not really afford the IP payments up front. RISC-V is a solution for one of the big components involved, which is the processor, which brings down the costs by a third. So, the DesignShare program addresses the problem involved with accessing third-party IPs.
SiFive uses RISC-V in its SoC cores. Image courtesy of SiFive.
AAC: How does DesignShare actually work?
SE: The idea is that we are going to be building the equivalent of an app store, but for IPs. IP vendors can have their IP added to the program, and it will be protected by SiFive since we will be the ones building the chip for the customer. We will provide the customer the IP models and so on, but the actual physical IP never leaves SiFive.
When we have our system up and running, the customer will be able to browse the IP library with an explanation of each IP with their specs, choose their processor, then the IPs needed for the application, and then can add their own IP. The customer then lets us know how many prototypes they need, and will pay SiFive some money for making the prototypes, but the money is not for the IPs; the money is for sharing the costs of mask and fabrication, so it’s very small compared to how much the customer would pay for doing it on their own.
After we deliver the prototype, the customer can then go to production, and before they sign the deal with us they will know exactly how much it will cost, including the IPs. They do not pay for the IPs until production.
AAC: What makes the DesignShare Program attractive to IP vendors?
SE: With the DesignShare program, the IP vendor does not have to deal with multiple customers at the same time: no sales support, no legal documents that you have to write with every customer, and the IP is secure with SiFive. That is why it’s attractive to IP vendors—because this opens up a new channel for them that they did not get exposed to before. They only deal with us as an aggregator, and there’s no overhead involved with sales, legal documentation, finding customers, and so on.
At the same time, we also lower the barrier for our customers to create prototypes. It isn’t until the customer’s idea takes off that they will pay for the IPs, and then we pay the IP vendors. It’s a win-win for SiFive, the IP vendors, and the customer.
AAC: How do you determine which IP vendors you partner with?
SE: If you look at any SoC in general, there is a foundation that everybody needs. For example, I/Os, memory—there is a basic foundation what we call Foundation IP that you have to have.
So, first we try to find a variety of IP vendors that have these basic IPs because this will be the foundation of any SoC. After that, we as a company will be defining vertical markets that we believe will be in high demand for a lot of new startups and small companies. These will be associated with the IoT, edge computing, and machine learning. Then we will be creating a library of IPs that will serve this market.
Currently, we are exploring how many IP vendors are interested and we are receiving great responses. We launched DesignShare two months ago and we have already announced eight vendors, which is almost one per week. Next year, we will probably continue at the same rate. So, we are going to be driving the IP library based on the vertical markets that we will be defining.
Flex Logix FPGA IP is part of the DesignShare program. Image courtesy of Flex Logix.
AAC: What's the demand like for support creating custom SoC design?
SE: We already have customers calling us. For example, we made an announcement with an embedded FPGA IP from Flex Logix, and a customer contacted us who wanted to combine an FPGA with a RISC-V processor—so we are exploring with them how many gates they need for the FPGA, the other IPs they’ll need around it, and so on.
The new concept of DesignShare plus open source for the core will lower the barrier for a lot of people who had no real dream of building such a system at a low cost compared to what they have done in the past. As you know, custom design has been going down over the years and the main reason is that nobody can afford it.
AAC: Having been involved in the industry for a while, are there any trends in the silicon world that you think will be important to watch out for in the near future?
SE: I think in the silicon world—I discussed this as a part of why we are going to be choosing certain vertical markets—there is a push in IoT for computing to be near the sensing part of the application. Right now, in order to do any computing, we still have to go to the cloud and datacenters. But there are a lot of applications that can not afford to send data to the cloud because maybe the Wi-Fi is down or the cellular service is down.
By moving the computing closer to sensors, you can avoid overwhelming the cloud or datacenters that maybe does not make sense to use in some cases. For example, Tesla cars have over 250 computing elements—imagine that you did not have edge computing or computing at every node. All these sensors would be sending data to the main processors in the car. In normal operation, nothing happens—but if all of the sensors keep sending data to the processor, it could overwhelm the processor.
If you start moving computing near the sensor, that means you only send the critical data to the processor. So you make intelligent decisions at the edge of the device, or near the sensor part of it, and that’s why you need some kind of computing element near the sensor. RISC-V could be one of the controllers with the IPs from DesignShare. So, in the future, computing is going to move closer to the edge.
Edge computing is becoming more important in applications such as autonomous driving. Image courtesy of Shutterstock.
AAC: Is there anything else you’d like our readers to know that maybe we haven’t discussed yet?
SE: At SiFive, we are trying to lower the barriers for innovation so that more people can innovate. We want to open innovation to the world. If you look at how many people can design, it’s only a handful of places on Earth that really have the capability or talent, so by making a lot of these IPs available for low cost or free, it will open up innovation to more people. SiFive will enable this.
AAC: Thank you for your time. Shafy. It was a pleasure to speak with you.