Dive Brief: - Honda Motor Co. and Texas-based semiconductor manufacturer Mythic will co-develop a system-on-chip for the automaker’s future software-defined vehicles, the automaker announced in a press release. As part of the project, Honda subsidary Honda R&D Co. will license Mythic’s analog, compute-in-memory processing technology. - Mythic’s technology performs calculations directly inside a SoC’s memory, rather than moving data to a centralized processor, which can significantly reduce power consumption, according to Honda. Mythic claims the approach is 100 times more energy-efficient than industry-standard GPUs, and can dramatically lower the costs of deploying advanced driver-assist and autonomous driving technology in future vehicles. - Honda confirmed to WardsAuto that this is a separate project with different “timelines and technical approaches” than its project to build a system-on-chip for SDVs with Renesas for future versions of Honda’s cancelled 0 Series EVs. “This initiative represents research and development targeting next-generation technologies beyond those examples,” it said in an emailed statement. Dive Insight: SoCs consolidate core computational and control tasks for various vehicle systems, including infotainment, propulsion, and safety-related and autonomous-driving systems. SoCs are widely considered more efficient and reduce the amount of wiring needed in a vehicle. Mythic’s analog compute-in-memory approach could fundamentally mean less shuffling of data. It uses a SoC’s memory as “tunable resistors” with inputs supplied as voltages and outputs as currents. Mythic also says it can strategically control the location of data in memory, which can improve efficiency when employing AI-powered neural networks for tasks such as image processing. Honda noted that it is “actively exploring neuromorphic SoC technology that draws inspiration from how the human brain works. Mythic claims its analog processing units can perform 120 million TOPS per watt of energy, which it says is 100 times more efficient than today’s top-performing GPUs performing memory transfers. In complex