Specifications
Models
Upbeat UP201 (compact / battery-driven devices) and UP301 (higher I/O and display capability)
Product Type
Always-on ultra-low-power AI microcontrollers for AI and IoT endpoints
CPU Cores
SiFive Essential RISC-V IP: E21 lightweight core and E34 performance core
Always-On Domain
E21 core in the AON domain manages continuous low-power sensing
Performance Domain
E34 core in the non-AON domain handles higher workloads
Operating Frequency
Up to 400 MHz
Compute
Up to 717 DMIPS
Power Approach
Near-threshold operation for ultra-low active and sleep power consumption
NPU 1
Handles complex AI workloads such as speaker ID, ECG and PPG analysis
NPU 2 (Always-On)
Dedicated always-on NPU for keyword spotting and gesture detection
DSP
Dedicated DSP acceleration for signal processing and sensor fusion
On-Chip Memory
Embedded SRAM with Upbeat's Error Detection and Correction architecture
External Memory
Expansion supported through SPI interfaces
Flash Interface
Quad SPI Flash interface on UP201
Mass Storage
Built-in SDIO 3.0 and eMMC 4.5.1 support on UP301
Graphics
Integrated GFX 2.5D GPU
Display — UP201
SPI (DBI-C) display interface
Display — UP301
2-lane MIPI-DSI-2 and QSPI display support
Camera
8-bit parallel camera interface on UP301
I/O — UP201
Up to 29x GPIO, 3x SPI, 3x UART, 10x I2C
I/O — UP301
Up to 100x GPIO, 6x SPI, 5x UART, 12x I2C
Target Applications
Wearables, drones, hearing aids, smartwatches and battery-powered sensor systems
Overview
Upbeat Technology, working with SiFive, developed the UP201 and UP301 as always-on AI microcontrollers for endpoints where the power budget, not the clock speed, is the binding constraint. Both parts use a dual-core RISC-V architecture: a SiFive E21 core sits in an always-on domain and manages continuous low-power sensing, while a SiFive E34 core in the main domain wakes for heavier work. Clocks run to 400 MHz and compute reaches 717 DMIPS, with near-threshold operation used to cut both active and sleep current.
Rather than one accelerator, the design carries two NPUs with different jobs. NPU1 handles demanding workloads such as speaker identification and ECG or PPG analysis, while an always-on NPU2 runs wake-word and gesture detection at the duty cycles those tasks need. A dedicated DSP covers signal processing and sensor fusion, and on-chip SRAM includes Upbeat's error detection and correction architecture — a reliability feature that matters in medical-adjacent devices.
The two parts diverge on I/O and display. The UP201 addresses compact battery devices with a quad SPI Flash interface, an SPI (DBI-C) display path, up to 29 GPIO, 3 SPI, 3 UART and 10 I2C. The UP301 steps up to SDIO 3.0 and eMMC 4.5.1 mass storage, a 2-lane MIPI-DSI-2 display interface with QSPI support, an 8-bit parallel camera interface, and up to 100 GPIO with 6 SPI, 5 UART and 12 I2C. Both include an integrated 2.5D GPU.
Key Benefits
Two NPUs with separate always-on and workload roles let wake-word detection stay live while the main NPU sleeps. SiFive E21 plus E34 RISC-V cores at 400 MHz and 717 DMIPS provide headroom for sensor fusion on a microcontroller power budget. Near-threshold operation targets multi-year battery life, and error-detecting and correcting SRAM plus a dedicated DSP support medical-adjacent signal-processing designs. The UP301 adds MIPI-DSI, camera input and up to 100 GPIO.
Applications
Smartwatches, hearing aids and wearables, always-on voice and gesture interfaces, ECG and PPG-based health monitoring, drone and battery-powered sensor systems, industrial IoT endpoints needing local wake-word detection.
Request a Quote — UPBEAT UP201 / UP301 — ALWAYS-ON RISC-V AI MCUS WITH DUAL NPU AND 2.5D GPU
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →