PARIS — With a body of engineering experience at LSI, Cisco, SiByte, Broadcom, PA Semi and Apple, Shailendra Desai is confident of his knowledge in SoC designs and what needs to be done.
Desai was senior engineering manager at Apple from 2007 to January 2013, where he cut his teeth on issues of interconnect architecture, as well as third-party and in-house IP integration. Desai said, “You can license individual IPs from various sources. But none of these IP vendors gave us a platform” to build and connect different IP blocks, in ways that might enable designers to optimize performance per watt for their SoCs.
Desai established Provino Technologies Inc. in 2015 to create a scalable platform for SoCs and IP subsystems. Provino hopes to help designers of SoCs for consumer, automotive and industrial applications where safety, security and energy efficiency are paramount, he explained.
For consumer products like iPod, iPhone and iPad, for example, it all comes down to how fast and how many derivatives one can pump out based on a single platform, the Provino CEO observed. The same holds true in the IoT SoC market. While acknowledging that IoT is “not a very well-defined space yet,” Desai cited an even greater demand for designing configurable and scalable SoCs quickly — on one platform.
Provino is still in its early phase. The company has neither talked to the press until now nor explained its technology in depth to industry analysts.
Provino nonetheless already has one unnamed Japanese automotive customer. Provino currently has 10 large customers evaluating its interconnect platform. “Not a single company has dropped thus far,” said Desai. Provino plans to go for series A funding in the first quarter of 2018.
The design challenges that Desai encountered with one client, while he was working as a consultant in 2013, foreshadowed similar problems across the board. His client designed an SoC that integrated various cores, including an FPGA. It was built by using Advanced eXtensible Interface (AXI). The company “couldn’t fit their design into an area they had. Meanwhile, AXI was creating a variety of problems such as congestions, latencies and performance,” Desai observed. “When I highlighted such issues in my report, the client said me, ‘Is that it? Can’t you solve the problem for us?’”
That client became Provino’s first customer. Provino today has 10 people working in California and 15 in India, and it’s hiring.
Interconnect: ‘Unsung hero’
During the interview, Desai confessed, “When I started Provino, I had no idea what others were doing” in the interconnect field.
As Mike Demler, a senior analyst at the Linley Group, put it, “Interconnect IP has been perhaps the unsung hero of the processor industry for at least 20 years. We just don’t hear about it as much as the latest CPU or GPU. ARM, Arteris, NetSpeed, and Sonics have a lot of expertise in this area.”
Demler explained, “Like any EDA tool and processor IP, there’s always room for improvement. I’d say that the application of machine learning, which NetSpeed is pioneering, could lead to big advances in SoC implementation.” After all, as he concluded, “Hooking up all the components of a billion-transistor chip is an incredibly complex task. Companies that can reduce design time and re-spin costs have an easy value proposition to win over customers.”
Desai told us that the multiple companies now pursuing the interconnect IP market gives him some comfort. “That means I am in the right market.” Indeed, Desai’s old colleagues at SiByte now hold positions at competing interconnect IP companies such as as Arteris and NetSpeed.
Building blocks of iFabric
So, what exactly is Provino’s interconnect platform?
Desai calls it “iFabric.” The list of iFabric’s building blocks is long, but it boils down to scalable architecture, a foundation for QoS (enabled by virtual channels), multi domain/topology support and integrated safety and security features.
First and foremost, “iFabric is a packetized and serialized bus with a concept of virtual channel,” Desai explained. iFabric provides multiple clocking options from server-class (high-speed) to consumer devices. Support for popular IP interfaces (AXI 3/4, APB and AHB) eases the integration of third-party and in-house IPs, he explained. Stressing its scalable architecture, he noted, “You can add or remove IPs easily so that you can design derivative products quickly.”
Next page: Issues and opportunities