Renesas Electronics Corp. has launched its first AI accelerated microcontroller (MCU), optimized for edge AI applications, delivering 256 gigabillion operations per second (GOPS) at 500 MHz. The new RA8P1 MCU Group targets voice, vision, and real-time analytics AI across a broad range of market segments. The company also announced its first comprehensive AI framework— Renesas Unified Heterogenous Model Integration (RUHMI)—for MCUs and MPUs, enabling efficient AI development and faster time to market.
Daryl Khoo, Renesas Electronics Corp.
“There are several trends in the market that drive demand for higher performance AI accelerated MCUs,” said Daryl Khoo, vice president of Embedded Processing Marketing Division at Renesas, during a presentation on Renesas’ AI microcontroller portfolio and gallium nitride (GaN) products.
“There is an increased demand for low power inference capability on edge devices that enable real time responses, lower latency, lower power consumption, lower cost, and higher security than a cloud-based solution,” he said.
Khoo also noted there is “a need for more integrated solutions that allow customers to reduce costs, and extend their investments in MCU-based designs without switching to power-hungry NPUs [neural processing units].”
The RA8P1 MCU is designed for diverse AI applications, in particular, multi-modal AI use cases that integrate vision and voice capabilities, and it is the next generation of the RA MCU family designed for the age of IoT, Khoo said.
Renesas’ first AI accelerated 32-bit MCU combines the Arm Cortex-M85 and Cortex-M33 cores with the Ethos-U55 NPU, delivering single- and dual-core RA8 MCUs that enable ultra-high 1-GHz CPU performance and built-in 256 GOPS of AI performance for demanding edge AI applications, he continued.
The MCUs establish a new performance level for MCUs by combining 1-GHz Arm Cortex-M85 and 250-MHz Cortex-M33 CPU cores with the 500-MHz Arm Ethos-U55 NPU. This combination delivers the highest CPU performance of over 7300 CoreMarks and AI performance of 256 GOPS at 500 MHz, according to the company.
Renesas’ RA8 Series MCUs, introduced in 2023, were the industry’s first based on the Arm Cortex-M85 processor that also included the Arm Helium technology for enhanced AI/ML performance.
The RA8P1 uses the Ethos-U55 NPU to offload the CPU for compute intensive operations in convolutional and recurrent neural networks (CNNs and RNNs), delivering up to 256 MACs per cycle that yield 256 GOPS performance at 500 MHz. The NPU supports commonly used networks such as DS-CNN, ResNet, Mobilenet, and TinyYolo. Depending on the neural network used, the Ethos-U55 provides up to 35× more inferences per second than the Cortex-M85 processor on its own.
“Customers are increasingly building multi-modal AI applications that combine voice and AI all in one device and require integration of hardware AI acceleration,” Khoo said.
To meet these applications, the RA8P1 packs dedicated peripherals for voice and vision AI as well as real-time analytics applications. These include a 16-bit camera interface that supports sensors up to five megapixels, enabling camera and vision AI applications. A separate MIPI CSI-2 interface offers a low pin-count interface with two lanes, each up to 720 Mbits/s. Other graphics peripherals include a graphics LCD controller supporting resolutions up to WXGA (1280 × 800), parallel RGB and MIPI-DSI display interfaces, a 2D Drawing engine, and a 32-bit external memory bus (SDRAM and CSC) interface.
Also included are multiple audio interfaces including I2S and PDM to support microphone inputs for voice AI applications. Other peripherals include Gigabit Ethernet and TSN Switch, XSPI (Octal SPI) with XIP and DOTF, SPI, I2C/I3C, SDHI, USBFS/HS, CAN-FD, SSI audio interfaces, a 16-bit ADC with S/H circuits, a DAC, comparators, a temperature sensor, and timers.
The RA8P1 offers both on-chip and external memory options for low latency neural network processing. The MCU includes 2-MB SRAM for storing intermediate activations or graphics frame buffers and 1 MB of on-chip MRAM for application code and storage of model weights or graphics assets.
High-speed external memory interfaces are available for larger models. System-in-package options with 4 or 8 MB of external flash in a single package will also be available in the future for more demanding AI applications.
The RA8P1 MCUs are manufactured on the 22ULL (22-nm ultra-low leakage) process from TSMC, which yields ultra-high performance and very low power consumption. This process also enables the use of MRAM in the new MCUs.
The advanced process technology meets high performance, high integration, and low power needs of industrial and AIoT customers, Khoo said, and the RA8P1 represents a significant leap forward in MCU technology by combining the most advanced Arm Cortex-M CPU with the advanced process node and emerging memory technologies such as the MRAM.
MRAM offers faster speed and higher retention and endurance than conventional flash memory technology while consuming less power, Khoo said.
“Memory technology advances are driven by the move to smaller geometries where embedded flash does not scale,” Renesas said.
Security features include the new Renesas Security IP (RSIP-E50D) with numerous cryptographic accelerators, including CHACHA20, Ed25519, NIST ECC curves up to 521 bits, enhanced RSA up to 4K, SHA2, and SHA3. Together with Arm TrustZone, this provides a comprehensive and fully integrated secure element-like functionality, Renesas said.
Other security features include root-of-trust and secure boot with first stage bootloader in immutable storage, and XSPI interfaces with decryption-on-the-fly that allow encrypted code images to be stored in external flash and decrypted on the fly.
Use cases
The RA8P1 MCU enables a variety of use cases in voice, vision, and real-time analytics. Voice AI use cases include keyword spotting, natural language understanding, and speech recognition with higher accuracy and lower power. These enable voice-activated user interfaces, smart speakers, and glass breakage detectors.
The RA8P1 also is suited for vision AI use cases such as image classification, object detection, and face or gesture recognition. These enable intelligence in traffic and driver monitoring applications, video doorbells, surveillance cameras, garage door openers, and door locks.
The AI MCU also can be used for real-time analytics use cases that enable motor failure detection, vibration analysis, fault diagnosis, and predictive maintenance applications.
“In all these diverse use cases, a huge uplift in inference processing performance—up to 36 times—can be seen with the use of Ethos-U55 NPU versus using the CPU core,” Khoo said.
Following the RA8P1, Renesas plans to release a 64-bit MPU with the Ethos-U55 NPU, which will further reinforce Renesas leadership in delivering truly scalable AI centric solutions, Khoo said.
(Source: Renesas Electronics Corp.)
Development tools
In addition to the RA8P1 MCUs, Renesas unveiled its RUHMI framework for MCUs and MPUs. It is the company’s first comprehensive AI framework for deploying the latest neural network models. It enables model optimization, quantization, graph compilation, and conversion, and generates efficient source code.
RUHMI provides native support for machine-learning AI frameworks such as TensorFlow Lite, Pytorch, and ONNX, and is integrated with Renesas’s own e2 Studio IDE to allow seamless AI development. It also provides the necessary tools, APIs, code generator, and runtime needed to deploy a pre-trained neural network, including ready-to-use application examples and models optimized for RA8P1.
Renesas also provides a wide range of tools and solutions for the RA8P1 MCUs, including the Flexible Software Package, evaluation kits, and development tools. FreeRTOS, Azure RTOS, and Zephyr RTOS are supported. Several software example projects and application notes are available.
In addition, Renesas offers several Winning Combinations designs using the new RA8P1 MCUs with compatible devices from its portfolio, including a Video Conferencing Camera with AI Capabilities, an AI Drawing Robot Arm, and an AI-Enabled Surveillance Camera. Partner solutions are available to support development with the RA8P1 MCUs, including a driver monitoring solution from Nota.AI and a traffic/pedestrian monitoring solution from Irida Labs. Other solutions can be found at the Renesas RA Partner Ecosystem Solutions Page.
The RA8P1 MCUs, housed in 224BGA and 289BGA packages, are available now. An RA8P1 Evaluation Kit is available. Samples and kits can be ordered either on the Renesas website or through distributors.
The post AI MCU sets new performance bar appeared first on Electronic Products.
