TOKYO — Renesas Electronics is pumped up. After lying low for a long time, the Japanese chip company is now brimming with confidence, claiming that it’s “ready to intersect” with the demand to produce highly automated vehicles in volume.
In store for Renesas’ roadmap is a yet-to-be-announced next-generation R-CAR SoC, designed for deep learning. Ready to go into advanced city driving autonomous vehicles in 2020, the new SoC will start sampling in 2019. It will deliver “more than 5 trillion operations per second (TOPS) at one watt,” Ryuji Omura, executive vice president at Renesas, revealed in an exclusive interview with EE Times in Tokyo last week. Omura is general manager responsible for the Japanese company’s automotive solution business unit.
The R-CAR SoC, if true, will double deep-learning performance efficiency compared to the Intel/Mobileye’s upcoming EyeQ5 SoC — which, according to Intel — will provide 24 TOPS at 10 watts.
Ryuji Omura, executive vice president at Renesas (photo: EE Times)
Until this surprising development, Renesas’ practiced pattern has been launching R-CAR SoCs primarily designed for in-vehicle infotainment applications and active safety systems. Renesas offered no hint about specific plans to design an R-CAR SoC for highly automated vehicles.
Indeed, unlike Nvidia and Intel/Mobileye, who have been battling to outshout each other about their self-driving expertise, Renesas has been conspicuously mute on the topic of robocar SoC platforms.
This reticence led some analysts to believe that Renesas might be more interested in ADAS than aggressively pursuing the nascent robocar market. Danny Kim, director and senior advisor at VSI Labs, for example, wondered if Renesas would consider TI and NXP as more of a direct competitor (with revenue largely earned from lower level safety systems) than Nvidia and Intel.
Nothing is further from the truth, according to Renesas. With the next-gen R-Car SoC underway, “We will be competitive in the highly automated vehicle SoC market,” said Omura.
Renesas’ game is to crack the robocar market at the right time with a much lower power and safety-conscious next-gen R-Car SoC, presenting itself as an alternative to Intel/Mobileye and Nvidia. Current lineup
Renesas believes that covering all of automotive computing tasks by only general purpose GPU calculation presents problems with overall performance and power. The basic concept behind the R-Car Autonomy Platform is to integrate various IP cores for different operations with high throughput internal bus, the company said.
Renesas’ two latest SoCs, both already available on the market, are R-Car H3 and R-Car V3M.
R-CAR H3 is designed as an automotive computing platform for the autonomous-driving era. The R-Car H3 provides cognitive computing capabilities and an enhanced performance that “can process large volumes of information from vehicle sensors accurately in real time, making it ideal for driving safety support systems,” according to Renesas. With the R-Car H3, car OEMs can run applications that require complex processing, such as obstacle detection, driver status recognition, hazard prediction and hazard avoidance. The same SoC can also handle in-vehicle infotainment systems.
Meanwhile, R-Car V3M is a high-performance image recognition SoC primarily aimed at front camera applications. It will serve surround-view systems or even lidars. It delivers low-power hardware acceleration for vision processing, and comes with a built-in image signal processor.
More specifically, the R-Car V3M processor includes two CPU ARM Cortex-A53, a dual lockstep ARM Cortex-R7, as well as IMP-X5+ (Renesas’ recognition engine optimized for interoperation with a CPU) and CNN specialized hardware blocks to accelerate computer vision and deep learning algorithms.
Next page: Road to autonomy