US- and Taiwan-based AI accelerator chip startup Kneron has entered the automotive market with strategic partnerships on aftermarket ADAS systems and a new strategic investor. With this move, the company intends to break into the notoriously difficult automotive sector.
Automotive is tough to break into for chip startups without years of reliability data available to prove their chips meet stringent requirements. On top of this, design cycles are long and startups without other revenue streams can’t afford to wait multiple years until vehicles go into production to start making revenues.
Kneron CEO Albert Liu told EE Times that Asian vehicle OEMs are just as difficult for startups to break into as their US and European counterparts.
“This is why we’re starting with the aftermarket,” he said. “As for our approach, we’ve been building revenue streams in the AIoT market and proving our products in vehicles. We’ve been building our reputation and record with channel markets.”
This is not Kneron’s first move in automotive. At the beginning of 2021, Kneron announced a strategic partnership with Taiwanese CEM Foxconn, where the companies will collaborate on AI use cases for Industry 4.0 and automotive through Foxconn’s MIH electric vehicle (EV) platform. MIH is an open software and hardware platform intended to lower the barriers to entry into the EV industry. Foxconn also made a strategic investment in Kneron at the time.
Strategic partnerships
Kneron’s two new strategic partnerships are both also with Taiwanese companies.
The first is with Weltrend, a fabless semiconductor company covering markets from signal processing to motor control to data security, and Elan/Avisonic, a human-machine interface company.
The three companies have been working on an ADAS system using Weltrend’s smart camera chip which handles optical flow algorithms, combined with Kneron’s KL520 AI accelerator chip which runs Elan/Avisonic’s AI image processing algorithm.
The Weltrend chip is responsible for the optical flow. Algorithms running on this chip consider a number of fixed points on an image and determine whether these points are moving closer or further from the camera in subsequent images. An alarm can be sounded if an object is within a bounded safety range of the vehicle, and/or the object is getting closer to the vehicle.
Adding Weltrend’s optical technology to the system increased detection accuracy by 70% in trials; the added detection accuracy decreases false alarms. The camera system can also be used to monitor driver behavior and detect traffic hazards.
The results of this collaboration will be used for aftermarket ADAS parking systems, such as cameras that consumers can place above their license plates or on their rear-view mirror.
“Our ADAS is not designed directly into a vehicle through an OEM, instead it reaches businesses and consumers through channel markets,” Liu said. “For businesses, it helps in fleet management – for public transport companies to monitor driver behavior or detect potential traffic hazards. Some of the projects most eagerly demanding these safety precautions include school bus projects.”
As part of this deal, Weltrend has also made a strategic investment in Kneron of undisclosed amount.
A second strategic partnership with camera solutions provider Otus has seen Otus’ 360° dashboard camera upgraded with the addition of the KL520. The camera system can now identify objects in the vehicle’s vicinity and alert the driver if necessary.
Industrial grade
The KL520 is Kneron’s first-generation AI accelerator chip, launched in May 2019. It offers 0.3 TOPS and consumes 0.5 W (0.6 TOPS/W). While the KL520 is not automotive qualified, it is an industrial grade chip, said Liu.
“Industrial grade is more than sufficient for [automotive] aftermarket applications based on all the current customer requests and specifications that we have seen,” he said.
Regarding reliability, the company currently has 6 months of test data from ongoing trials of KL520-equipped aftermarket systems installed in vehicles on the road.