Neural engine IP - Tiny and Mighty

Overview

Small, low-power dedicated AI engines are essential for home appliances, security cameras, and always-on smartphone features. Customized for specific use cases, Origin™ E1 delivers targeted low-power performance and requires little to no external memory.

The Origin E1 NPUs are individually customized to various neural networks commonly deployed in edge devices, including home appliances, smartphones, and security cameras. For products like these that require dedicated AI processing that minimizes power consumption, silicon area, and system cost, E1 cores offer the lowest power consumption and area in a 1 TOPS engine.

Power-Sipping, Always-Sensing AI

Always-sensing cameras continuously sample and analyze visual data to identify specific triggers relevant to the user experience. They enable a seamless, more natural user experience. However, always-sensing data requires specialized AI processing due to the quantity and complexity of data generated. OEMs are turning to specialized AI engines like Expedera’s LittleNPU. The LittleNPU is optimized to process the low-power, high-quality neural networks used by leading OEMs in always-sensing applications. It runs at low power—often as low as 10-20mW—and keeps all camera data securely within the LittleNPU subsystem to preserve user privacy.

Innovative Architecture

The Origin E1 neural engines use Expedera’s unique packet-based architecture, which enables parallel execution across multiple layers, achieving better resource utilization and deterministic performance. This innovative approach significantly increases performance while lowering power, area, and latency.

Specifications

Compute Capacity	0.5K INT8 MACs
Multi-tasking	Run Simultaneous Jobs
Power Efficiency	18 TOPS/W effective; no pruning, sparsity or compression required (though supported)
Example Networks Supported	MobileNet, EfficientNet, NanoDet, PicoDet, Inception V3, RNN-T, MobileNet SSD, BERT, FSR CNN, CPN, CenterNet, Unet, YOLO V3, ShuffleNet2, others
Layer Support	Standard NN functions, including Conv, Deconv, FC, Activations, Reshape, Concat, Elementwise, Pooling, Softmax, others.
Data types	INT4/INT8/INT10/INT12/INT16 Activations/Weights
Quantization	Channel-wise Quantization (TFLite Specification) Software toolchain supports Expedera, customer-supplied, or third-party quantization
Latency	Deterministic performance guarantees, no back pressure
Frameworks	TensorFlow, TFlite, ONNX, others supported

Key Features

Choose the Features You Need: Customization brings many advantages, including increased performance, lower latency, reduced power consumption, and eliminating dark silicon waste. Expedera works with customers to understand their use case(s), PPA goals, and deployment needs during their design stage. Using this information, we configure Origin IP to create a customized solution that perfectly fits the application.
Market-Leading 18 TOPS/W: Sustained power efficiency is key to successful AI deployments. Continually cited as one of the most power-efficient architectures in the market, Origin NPU IP achieves a market-leading, sustained 18 TOPS/W.
Efficient Resource Utilization: Origin IP scales from GOPS to 128 TOPS in a single core. The architecture eliminates the memory sharing, security, and area penalty issues faced by lower-performing, tiled AI accelerator engines. Origin NPUs achieve sustained utilization averaging 80%—compared to the 20-40% industry norm—avoiding dark silicon waste.
Full TVM-Based Software Stack: Origin uses a TVM-based full software stack. TVM is widely trusted and used by OEMs worldwide. This easy-to-use software allows the importing of trained networks and provides various quantization options, automatic completion, compilation, estimator and profiling tools. It also supports multi-job APIs.
Successfully Deployed in 10M Devices: Quality is key to any successful product. Origin IP has successfully deployed in over 10 million consumer devices, with designs in multiple leading-edge nodes.

Benefits

1 TOPS performance
Support for standard, custom, and proprietary neural networks
Performance efficiencies up to 18 TOPS/Watt
Full software stack provided, including compiler, estimator, scheduler, and quantizer
Runs CNN, RNN, DNN, LSTM, and other network types
Delivered as Soft IP (RTL) or GDS

Block Diagram

Applications

Home Appliances
Smartphone
Consumer Electronics
Handheld devices

Technical Specifications

Maturity

In production

Availability

Production

Request Info