Renesas has introduced a brand new single- or uneven dual-core microcontroller vary, focusing on energy-efficient but high-performance synthetic intelligence and machine studying (AI and ML) on the edge: the RA8P1.
“There’s explosive progress in demand for high-performance edge AIoT [Artificial Intelligence of Things] purposes,” says Renesas’ Daryl Khoo in help of the corporate’s newest microcontroller design. “We’re thrilled to introduce what we consider are the perfect MCUs [Microcontroller Units] to deal with this development. The RA8P1 gadgets showcase our know-how and market experience and spotlight the sturdy partnerships we’ve constructed throughout the business. Clients are wanting to make use of these new MCUs in a number of AI purposes.”
Renesas has introduced a beefy new microcontroller focusing on edge-AI workloads: the RA8P1. (📷: Renesas)
The RA8P1 household comes with a main Arm Cortex-M85 core operating at a speedy 1GHz, plus an optionally available secondary Cortex-M33 core operating at as much as 250MHz. Extra importantly for Renesas’ imaginative and prescient because the driving pressure behind an edge AI push is the presence of an Arm Ethos-U55 neural coprocessor, which may ship as much as 256 giga-operations per second (GOPS) of minimum-precision compute at 500MHz — delivering, in real-world phrases, as much as 35 instances the inference efficiency of the Cortex-M85 core by itself, the corporate says.
Manufactured at Taiwan Semiconductor (TSMC) on the corporate’s energy-efficient 22nm Extremely-Low Leakage (22ULL) course of node, the brand new chips embrace both 512kB of 1MB of on-chip magneto-resistive RAM (MRAM) instead of the same old program flash to carry the facility draw down nonetheless additional. Every chip additionally consists of 2MB of static RAM (SRAM) with error correction code (ECC), 32kB instruction and knowledge caches per core, and the choice of including 4MB or 8MB of exterior flash in a system-in-package (SIP) single-chip sandwich.
The corporate has launched a trio of reference designs constructed across the half, together with this robotic drawing arm. (📷: Renesas)
Peripherals embrace a graphics controller supporting outputs as much as 1280×800 (WXGA), parallel RGB and MIPI Show Serial Interface (DSI) outputs, a 2D drawing engine, two-lane MIPI Digicam Serial Interface 2 (CSI-2) and 16-bit CEU digital camera inputs, gigabit Ethernet with Time Delicate Networking (TSN) help, octal-SPI with execute-in-place (XIP) help, SPI, I2C, I3C, SDHI, USB Full Velocity and Excessive Velocity, CAN FD, PDM and SSI audio interfaces, a digital to analog converter (DAC), 16-bit analog to digital converter (ADC), and a cryptographic subsystem together with Arm TrustZone help.
Concurrently the {hardware}, Renesas has additionally introduced the Renesas Unified Heterogeneous Mannequin Integration (RUHMI) framework. This, the corporate says, permits for the deployment of neural community fashions — together with native help for TensorFlow Lite, PyTorch, and ONNX — with full optimization for RA8P1 components. For these seeking to get began shortly, the software program additionally consists of all the pieces required to deploy pre-trained neural networks together with a choice of fashions and software examples written for the RA8P1.
An analysis equipment places the microcontroller on a breakout board with a bundled digital camera module and show. (📷: Renesas)
As is common for the corporate, Renesas is providing a variety of “Successful Combos” reference designs that pair the brand new RA8P1 microcontrollers with the corporate’s different elements to construct a choice of gadgets: an AI-enhanced video conferencing digital camera, an AI-powered robotic drawing arm, and an AI-based surveillance digital camera.
Extra info on the RA8P1 chips is out there on the official product web page; Renesas has additionally launched the EK-RA8P1 Analysis Equipment improvement board, which pairs the half with a 7″ 1024×600 parallel LCD panel and a 5 megapixel digital camera module, priced at $198.