Flex Logix Technologies has closed a $55 million series D funding round, which it will use to scale up its artificial intelligence (AI) inferencing part of the business. This round, led by Mithril Capital Management plus existing investors Lux Capital, Eclipse Ventures and the Tate Family Trust, takes the company’s total funding since being founded in 2014 to $82 million.
In an interview with EE Times, Geoff Tate, CEO and co-founder of Flex Logix, said that while the company’s original eFPGA business was already a commercial success, the investment would help grow its inferencing business. “Our inferencing chip, InferX X1 is sampling to customers at the moment. The funding will enable us to ramp up production of this chip and build a new sales team to focus on this side of the business.” He said the company plans to hire some 50 people over the next nine months, doubling in size from its current 50 staff.
“Our eFPGA business is already cash positive. Our initial engagement in the early years was government, particularly DARPA (Defense Advanced Research Projects Agency). That got us lots of projects, and also commercial customers, such as Dialog Semiconductor for their configurable power management ICs (PMICs),” said Tate. Dialog licensed Flex Logix’ eFPGA and its compiler at the end of 2019, to enable configurability of Dialog’s future PMICs.
Flex Logix’s inference architecture is optimized for low latency operation required by edge megapixel vision applications. It combines numerous 1-dimensional tensor processors with reconfigurable, high bandwidth, non-blocking interconnect that enables each layer of the neural network model to be configured for maximum utilization, resulting in very high performance at lower cost and power. The connections between compute and memory are reconfigured in millionths of a second as the model is processed. This architecture is the basis of Flex Logix’s InferX X1 edge inference accelerator which is now running YOLOv3 object detection and sampling to lead customers.
Tate said this programmable interconnect architecture evolved from the eFPGA architecture, explaining the background. “Just over 2 years ago a customer was asking for a suggestion on how to implement inferencing on FPGA. My co-founder came up with a new architecture, InferX. In many FPGAs, 80% of it is used for programmable interconnect, not programmable logic. That same programmable interconnect is used for inferencing, and it’s non-blocking, which addresses customer needs to address latency issues.”
Earlier this month, Flex Logic said it expanded its license agreement with DARPA for its EFLX eFPGA technology. The agreement allows DARPA researchers to maintain all the benefits of hard-wired circuitry, while also adding the capability to change the circuit after tape-out if they need to modify, update or fix a functional bug in their algorithm or accelerator. Tate said, “Since partnering with DARPA in 2017, Flex Logix has provided its eFPGA to many U.S. government funded programs which need the flexibility and reprogrammability of eFPGA in chips at more advanced nodes. Because researchers often only get funding to tape-out a chip once, it makes strategic sense to also include EFLX and the EFLX Compiler to the Toolbox Initiative in support of DARPA’s research programs.”
The latest investment is led by Mithril Capital, who recently also led the series B investment in Nuvia (which was recently acquired by Qualcomm for $1.4 billion). Mithril’s managing general partner and founder, Ajay Royan, said, “We are impressed with the very high inference-throughput/$ architecture that Flex Logix has developed based on unique intellectual property. This technology advantage positions Flex Logix for rapid growth in edge enterprise inference in applications such as medical, retail, industrial, robotics and more. It is even more impressive that they have done this with so little capital and at the same time built a cash-flow positive eFPGA business with large growth potential as SoC designers look to incorporate reconfigurability into their communications and data center ICs.”
Tate said, “Mithril’s strong support, combined with the continued backing from our existing investors, will allow us to further build out our software, engineering and customer support teams and accelerate the availability of our hardware and software for edge enterprise applications. Our InferX X1 chips and boards will be available for mass production in mid-2021 along with availability of our InferX Inference compiler, which takes in Tensorflow Lite and ONNX neural network models and generates the code to run InferX X1 without the detailed programming other solutions require.”
Investors are indeed relying on Tate’s track record. From managing AMD’s microprocessor and logic group with over 500 people to leading his first startup, Rambus, from four people and $2 million in equity to a NASDAQ IPO and multi-billion dollar market cap. Peter Hebert, co-founder and managing Partner of Lux Capital. “We all have confidence Geoff can do it again, building Flex Logix into an industry-defining, stand-alone public company.”
So now with the series D, is there an IPO on the cards next, we asked Tate? His response: “The plan is to go public. But we don’t know when, it could be a few years out yet.”
The post Flex Logix Raises $55M to Scale AI Inferencing Business appeared first on EETimes.