The chip is intended for high performance image acquisition and supports convolutional neural network vision workloads for object identification.
German company Dream Chip Technologies has introduced at the Mobile World Congress the first chip to be manufactured in the 22FDX 22nm FDSOI manufacturing process of Globalfoundries.
The chip is an advanced driver assistance system (ADAS) computer vision SoC developed with the support of the European Things2Do project in cooperation with ARM, Arteris, Cadence, Globalfoundries and Invecas. The chip was manufactured at Fab 1 in Dresden, Germany, where Globalfoundries hosts its FDSOI manufacturing.
Dream Chip is a design service company specialising in the development of ASICs, FPGAs, embedded software and systems with a strong application focus on automotive vision systems. The company is the result of more than 20 years of endeavour having been founded as Sican by the government of lower Saxony in 1990. In 2000, it was acquired by Infineon Technologies and operated as Sci-worx. It was then bought by Silicon Image before a management buy-out in 2010 and a name change to Dream Chip Technologies.
The chip the team has developed is intended for high performance image acquisition and supports convolutional neural network (CNN) vision workloads for object detection and identification. It supports road-sign detection, lane departure warning, driver distraction warning, blind spot detection, surround vision, park assist, pedestrian detection, cruise control and emergency braking.
The selection of the 22nm FDSOI process is intended to allow all this at low power, according to the company.
The chip includes a proprietary image signal processing pipeline—DSP for image processing was one of the strengths of Sican—as well as multiple licensed-in cores. These include quad Cortex-A53 cores, Tensilica Vision P6 DSPs and dual Cortex-R5 to provide functional safety compliant to ISO26262. The cores are interconnected using a FlexNoC network-on-chip licensed from Arteris. Invecas has supplied foundation IP such as LPDDR4 interface, PLL, thermal sensor and process monitor.
"The project not only shows the strength of the European semiconductor industry, but also an ability to collaborate efficiently to provide technology needed urgently by the industry to power advanced automated driving solutions," said Jens Benndorf, managing director and co-founder of DCT, in a statement.