Chip suppliers, startups team up to advance edge AI for biometric applications
Chip suppliers are pairing up to offer cutting-edge technology for facial and voice recognition as emerging companies such as Syntiant and Xailient move from the concept stage to production for edge AI processors. New developments show edge AI technology being used to enable biometric use cases that enable huge performance gains with low power consumption—low enough for standalone (unplugged) applications in factories and homes.
Renesas has teamed with Syntiant to jointly develop a voice-controlled “multimodal” AI processor, with multimodal referring to the combination of voice and object recognition capabilities. The companies expect the chips to be used in smart appliances or retail settings where cashierless checkout would require face and object detection, for example.
The joint solution combines the Renesas RZ/V series vision microprocessor with Syntiant’s NDP120 for neural network processing. The reference design enables always-on functionality along with voice-triggered activation from standby mode to perform object recognition, facial recognition, and other vision-based tasks, according to the companies. This means security cameras and other systems can track objects (including people) while letting users trigger actions with voice cues for activation or operation of robotic arms in manufacturing settings, for example.
“We anticipate that demand for multimodal systems that use multiple streams of input information – both image and voice – will increase moving forward as a way to improve both ease of use and safety,” said Hiroto Nitta, senior vice president and head of SoC Business in the IoT and Infrastructure Business Unit at Renesas, in a prepared statement.
Renesas said that the reference design for the new multimodal AI solution is currently available.
In related news, Maxim, a $2.2B provider of integrated circuits and chip designs, said it was integrating technology from Xailient to offer a chip with super-fast biometric face detection while using less power than other traditional chip designs. The MAX78000 is an ultra-low power neural-network microcontroller; when using Xailient’s Detectum neural network technology, Maxim claims that the chip can detect and localize faces in video and images in 12 milliseconds per inference while using 250 times less power than conventional chip designs.
The result, according to the companies, is that battery powered edge AI systems such as security cameras for home and industrial or retail applications that are just as accurate as other cameras. In some applications, the operating time of a coin-sized battery powered using a hybrid edge/cloud applications could last for many years in the field.
The MAX78000 is currently available at Maxim Integrated’s website and distributors while the Detectum neural network, series models, tools, and services are available from Xailient.
The pair of announcements show how fast the edge AI chip market is developing, moving from concept to shipping products in a short span of time as equipment manufacturers seek to design net-generation security cameras and control systems. The involvement of large companies such as Renesas and Maxim reinforce the aggressive growth targets for edge AI.
Market research company Omdia forecasts that global AI edge chipset revenue will grow from $7.7 billion in 2019 to $51.9 billion by 2025, while Deloitte has predicted edge AI chip unit shipments will exceed 1.5 billion by 2024.