Nextchip will supplying its advanced driver assistance system (ADAS) system-on-a-chip (SoC) to a Japanese tier-1 supplier.
The company was preparing to manufacture the Apache 4 for supply, Nextchip chief marketing officer (CMO) Yoo Young-jun told TheElec.
It marks the first time that the company is supplying edge ADAS SoC to a Japanese customer.
Nextchip also launched the prototype of Apache 5, which has deep learning technology applied, late last year, the CMO said. Customer demo of the newer chip wrapped up recently, he added.
ADAS SoC comes in central processing method and distributed processing method (also called edge processing).
Central processing method processes every data on the electronic control unit. This requires large computing power and consumers more power.
The distributed method has the sensors execute the data detection function, then sends the sorted data to the ECU. This takes less power compared to central processing.
Apache 4 consists of a CPU, digital signal processor and an image signal processor.
Apache 5 adds a neutral processing unit to this set up. The NPU had convolution neural network algorithm applied which applies for 1.6TOPS and 800MHz, Nextchip said.
Apache 4 is manufactured by Samsung Foundry with 28-nanometer process. Apache 5 is made with the 14-nanometer process by the South Korean tech giant.
Nextchip formed a special team to develop the deep learning algorithm on the Apache 5, Yoo said. Customers can choose to use the algorithm or their own or both, the CMO said.
The chip can recognize road lanes, road signs, pedestrians in the dark and children under 80cm tall, he said. It can support Level 3 to Level 4 autonomous driving, he added.
Nextchip is expanding the supply of ISP with analog high definition technology. The chip has been supplied to a South Korean tier-1 auto-component supplier which provided it to Hyundai Avante and Hyundai Santa Fe.
Nextchip was founded in 1997 as a developer CCTV ISP. It split off from parent company NC& last year. It is planning to go public in 2022.