NXP Semiconductors’ recent acquisition of Kinara was the latest in a string of edge AI acquisitions, and there could be more to come as more companies look to push AI processing to edge devices via neural processing units (NPUs), FPGAs, and other chips and capabilities.
The NXP-Kinara deal in particular put NPUs in the spotlight. “It is clear that no IoT or automotive or industrial chipmaker can put products in the market without an NPU,” said Jack Gold, president and principal analyst at J. Gold Associates, in an email exchange with Fierce Electronics. “Auto, robotics, industrial controllers, and even consumer devices will all move towards embedded AI.”
NXP’s $307 million agreement to buy Kinara, announced earlier this month, followed Analog Devices Inc.’s acquisition of FPGA player Flex Logix late last year. Amazon also had edge AI advances in mind last year when it acquired early AI chipmaker startup Perceive, and there have been some smaller-scale edge AI acquisitions as well. Gold added that more consolidation could be coming to an edge AI sector that has become overstuffed with startups in the last few years.
“There will no doubt be other such acquisitions as there are too many AI startups in chip design,” he said.
For NXP, the Kinara acquisition beefs pre-existing, in-house NPU efforts. The company announced its Neutron NPU device in early 2023, but as different devices continue to be positioned for different kinds of AI workloads, Gold said the Kinara technology NXP is acquiring will strengthen and deepen its edge AI story.
“With NXP’s focus on the industrial and auto sectors, they need a robust NPU story to tell, so bringing on other tech, especially as it relates to those markets, is a good move… I think this is also beneficial to Kinara. Trying to build a market for their products in market areas that can take two, three, or four years to get to volume would be a difficult proposition for them in obtaining sales momentum and funding growth.”
Rutger Vrijen, senior VP of strategy at NXP, during a recent media briefing following the announcement of the Kinara deal, echoed that sentiment, explaining that NXP can help Kinara to scale up from its early success of having sold about 500,000 of its NPUs. “Many of the customers they work with are at the ramp stage of shipping so we expect that 500K number to grow quickly as a result of joining forces with NXP.”
NXP’s expects that growth to come from edge AI deployment opportunities in industrial and consumer IoT, “where we see combinations of [AI] models for vision, voice, security, anomaly detection, often in multimodal configurations, where a single NPU and system can drive, for example, both a vision and a language model to provide comprehensive intelligence of an environment or centralized decision making,” Vrijen said.
He added that automotive applications where Kinara’s NPUs could have an impact, outside of ADAS, include in-cabin monitoring and in-vehicle infotainment.
Vrijen also waxed on about the broader implications for greater AI processing at the edge, saying that it could be “really a transformative force. We don't always want to rely on a cloud and a connection to it to drive our information, interpretation, and decision making. This can now happen at the edge with a highly energy-efficient solution… that provides human-like interactions with the use of generative AI systems that provide contextual understanding… for example, classifying an image, sending that into a large language model, getting information about what's going on in that image or video, and driving smarter decision making as a result. But it all has to happen with the right level of data privacy, cost effectiveness, scalability and energy efficiency for it to be feasible at the edge, and that's really where the acquisition of Kinara comes in.”