RA8P1 microcontroller chip image from Renasas
Japanese semiconductor firm Renesas Electronics has unveiled the RA8P1, positioning it as a breakthrough microcontroller that integrates gigahertz-class processing, dedicated AI acceleration hardware and magnetoresistive RAM (MRAM) on a single 22-nanometer chip. The chip includes graphics capabilities, multiple connectivity interfaces and advanced security features typically associated with higher-end processors.
In a press briefing, Renesas billed the RA8P1 as a significant leap forward in microcontroller technology. Daryl Khoo, vice president of Renesas’ embedded processing marketing division described the chip as “the next generation of Renesas RA microcontroller family, designed for the age of AIoT.” While the term “AIoT” has been tossed around for years, representing as it does the convergence of artificial intelligence and Internet of Things, it’s only now becoming practical as chips gain the power to run AI models locally without relying on cloud processing. “If we truly want to enable intelligence at the extreme edge and endpoints in the network, it must be done in a way that is highly efficient, responsive in real-time, and cost-effective,” Khoo said.
Renesas tackles the “Moore’s Law of Power” with new GaN FETs
Renesas’ new GaN FETs target diverse high-power applications including battery energy storage, e-mobility charging, and AI data center infrastructure.
Beyond edge AI processing, Renesas is also addressing the power infrastructure needed to support these compute-intensive applications. Renesas has introduced 650V Gen IV Plus SuperGaN Field-Effect Transistors (FETs) that cut switching losses by 10-30% compared to silicon carbide technology. The devices feature 14% smaller die size and 14% lower on-resistance than previous generations. “Silicon, an excellent material, has reached its physical limits in power conversion. And that is where GaN enters,” said Primit Parikh, VP of Renesas’ GaN Business Division, referring to gallium nitride, a rising star in the semiconductor world. “GaN is lower loss versus silicon carbide at any frequency. This gap widens at higher frequency.” The strategic focus on GaN reflects broader market dynamics. “Given the market dynamics over the last two years, as discussed in our capital market, Renesas made a strategic decision to temporarily suspend the silicon carbide and IGBTs, and hence now focusing on the discrete side, on gallium nitride and MOSFETs. This pivot allows us to capitalize on key areas like data center, power infrastructure growth,” Parikh explained. The new devices target multi-kilowatt applications in server power supplies, e-mobility charging systems and industrial power conversion.
To achieve this vision of local AI processing, Renesas integrated three key technologies in the RA8P1. The 22 nm manufacturing process, a relatively advanced node for MCUs, though mature by standards of companies like TSMC, which produces cutting-edge chips for Apple and NVIDIA at 3 nm and 5 nm nodes, enables the gigahertz speeds. MRAM replaces traditional flash memory. Arm’s Ethos-U55 Neural Processing Unit targets the TinyML market. When asked why Renesas didn’t use Arm’s newer Ethos-U85, Khoo acknowledged it was ‘a new development’ they’re evaluating for future products, a reminder that even cutting-edge chips face time-to-market tradeoffs.
The 22 nm process does offer clear advantages over the 40 nm and 55 nm nodes typical in MCUs. The 1 MB of on-chip MRAM provides faster write speeds and higher endurance than embedded flash. Security wasn’t an afterthought. The chip includes Renesas’ RSIP-E50D hardware security module with cryptographic accelerators and secure boot capabilities.
Renesas has invested in the developer experience. “One of the challenges that we see in the AI emerging market is customers’ familiarity with AI development,” said Khoo. “Today, we introduced RUHMI, where we actually integrated the AI development workflow into our embedded systems development workflow. So in one common workflow, the customers can then develop AI solutions quicker and faster.”
RUHMI (Renesas Unified Heterogenous Model Integration) is integrated into the company’s e² studio IDE. The framework generates optimized neural network models and supports TensorFlow Lite and PyTorch, addressing the real barrier to AI adoption at the edge.
This multipronged innovation in both edge AI and power infrastructure is the result of a deliberate and significant R&D strategy. The company invested 249.6 billion yen (approximately $1.72 billion) in R&D in 2024, representing 18.5% of total revenue.
