Another day, another semiconductor company talking about its own take on AI processing at the edge.
Earlier this week, Ambarella unveiled its solutions for on-camera AI processing. Next up is Quadric, a five-year-old company that previously created agricultural robots that didn’t quite take root, now unveiling a general-purpose parallel processor platform targeting on-device AI for computer vision and image processing.
Quadric is claiming a differentiator in its hybrid processor architecture, which can handle classic image processing algorithms and any newer AI and machine learning algorithms, according to Daniel Firu, chief product officer at Quadric.
Firu told Fierce Electronics that back when Quadric’s founders were developing agricultural robots to monitor soil and other aspects of farms they realized they “weren’t satisfied with the compute available to us. We used the Nvidia [Jetson] TX1 GPU, and our algorithms just saturated it. So we decided to build the one we would want to use as developers.”
The Quadric q16 processor architecture is based on a hybrid data-flow and Von Neumann machine that enables high-performance on-device computing for demanding workloads including neural networks, machine learning, computer vision and basic linear algebra subprograms (BLAS), according to a Quadric press release. Firu described it as a more “unified” platform than the heterogeneous systems composed of CPUs, GPGPUs, FPGAs, and AI chips that can vex developers with their complexity, power requirements and software integration needs.
The result is a processor architecture that is optimized not only for edge AI, but on-device AI processing, an application Ambarella also is targeting. “This is a solution that can go anywhere you want to take sensor info, process it and make a decision on it,” Firu said. That means cameras and other computer vision equipment for communications, medical and automotive applications (automaker Denso is an investor), among others.
Edge AI and on-device AI processing seem destined to become more active arenas for all semiconductor developers. Jack Gold, principal analyst with J. Gold Associates, told Fierce Electronics, “While AI at the edge is not yet a huge market, it will become one over the next 2-3 years as workloads get dispersed to be closer to the end points and distributed to do much more preprocessing before sending to the cloud data centers. So it’s definitely a growth market. And in the short term a lot of that growth will be in image processing.”
However, with increasing numbers of semiconductor firms aiming at these opportunities (Xilinx was another recently hyping edge AI) there won’t be room for everyone to succeed. “So the bottom line is, over the next 2-3 years, we’re going to see a lot of shakeout in this market as some of the myriad of players either get bought by the bigger players, or fade away. That’s pretty typical of almost any emerging market, including AI and AI at the Edge.”