top of page
World’s Best AI SoC for On-Device AI
REGULUS, the world's most efficient AI SoC, empowers your devices with unparalleled on-device inference capabilities. Leveraging its advanced AI engine and high-performance CPU, REGULUS enables your devices to execute complex inference tasks with unrivaled efficiency and cost-effectiveness. Break free from cloud dependencies and unlock the full potential of your robotic, drone, and AI camera applications with Regulus’s ability to process high-resolution images with minimal latency and maximum accuracy.
.png)
Seamlessly Processes 4K 60FPS UHD Video with Ultra-Low Latency
REGULUS is powered by an advanced AI engine and state-of-the-art ISP, REGULUS delivers world-class image processing capabilities. Leveraging High Dynamic Range (HDR), REGULUS ensures accurate image processing even in low-light conditions. REGULUS preserves critical details while effectively eliminating noise with 3D Noise Reduction (3DNR), enabling precise AI functionality. Coupled with multi-standard video codec support, REGULUS empowers your devices to effortlessly handle high-resolution, high-framerate video processing, redefining the boundaries of visual AI.
REGULUS
Leading AI SoC for on-device AI
%201.png)
REGULUS
Ultra-Compact AI System-on-Chip
REGULUS is a compact AI System-on-Chip(SoC) with a 10 TOPS computing power at under 3W of power, optimized for higher-performance on-device AI use cases in power- and space-constrained applications such as drones and security systems.
REGULUS comes pre-equipped on a system-on-module (SoM) board, making it a fully functioning AI PC in itself.
Product specs
NPU
10 TOPS
CPU
Quad-core ARM Cortex-A53,
ARM Cortex-M0+
Power (TDP)
3 W
Dimensions
17 mm ✕ 17 mm
Memory
Up to 8 GB (DDR4, LPDDR4, 4x)
Memory Speed
4266 Mbps
Storage
SD, SDIO, eMMC 5.1
Image Processing
ISP (8M @ 60 FPS), Dual DVP
Key features: Break free from GPU dependency today
Cost effective
Most cost-effective options for engineers prioritizing high price-performance ratio.
Scalable
Scalable NPU architecture supporting 1 to 100+ TOPS.
High accuracy
World-leading lightweight technology maintains 99 % accuracy of existing models.
Easy to use
User-friendly full-stack SDK that supports major ML frameworks including Tensorflow, Pytorch, ONNX.
Eco friendly
Eco-friendly solution that maximizes data reuse and minimizes memory access for high energy efficiency.
Programmable
Supports more than 300 deep learning models, including SOTA models, with excellent performance.
bottom of page