Highly scalable inference NPU IP for next-gen AI applications

ENLIGHT Pro is a high-performance, scalable NPU IP designed for edge AI applications, including automotive and cameras. It supports Transformer models and delivers 4,096 INT8 MACs/cycle, with performance scalable from 8 TOPS to hundreds of TOPS. Single-, dual-, and quad-core configurations are available, along with multiple data types and tensor shape transformation operations.
 

ENLIGHT Pro incorporates a RISC-V CPU vector extension with custom instructions and supports multiple task mappings, including multiple models, data parallelism, and tensor parallelism. The ENLIGHT SDK supports widely used network formats including ONNX (PyTorch), TFLite (TensorFlow), and CFG (Darknet), providing software tools for network conversion and NPU deployment.

Toolkit Overview

• NN Converter: 
  • Converts a network file into internal network format (.enlight)​
  • Supports ONNX (PyTorch), TF-Lite, and CFG (Darknet)
• ​​NN Quantizer
  • Generates ​quantized network: float to 4-/8-bit integer
  • Supports per-layer quantization of activation and per-channel quantization of weight 
• ​NN Simulator
  • Evaluates full precision network and quantized network​
  • Estimates accuracy loss due to quantization
• ​​NN Compiler
  • Generates NPU handling code for target architecture and network​

• Mixed-Precision Computation (INT8, INT16, FP16): Achieving accuracy while preserving power, performance, and area (PPA) efficiencies
• Deep Neural Network (DNN)-optimized Vector Engine: Custom instructions for Softmax and local storage access & enhanced adaptability for future DNNs
• Scale-out w/ Multi-core: Greater performance by parallel processing of DNN layers
• Modern DNN Algorithm Support: Transformer architecture, depth-wise convolution, feature pyramid network (FPN), etc.
• High-level Inter-layer Optimization: Optimized layer grouping and scheduling to minimize DRAM traffic from intermediate data
• DNN-layers Parallelization: Effective multi-core utilization for elevated performance & optimized core-to-core data transfer
• Automated Quantization Flow: Minimization of quantization loss through mixed-precision computation

ENLIGHT Pro Hardware Key Advantages

• Mixed-Precision Computation (INT8, INT16, FP16): Achieving accuracy while preserving power, performance, and area (PPA) efficiencies
• Deep Neural Network (DNN)-optimized Vector Engine: Custom instructions for Softmax and local storage access & enhanced adaptability for future DNNs
• Scale-out w/ Multi-core: Greater performance by parallel processing of DNN layers
• Modern DNN Algorithm Support: Transformer architecture, depth-wise convolution, feature pyramid network (FPN), etc.

ENLIGHT Pro Software Key Advantages

• High-level Inter-layer Optimization: Optimized layer grouping and scheduling to minimize DRAM traffic from intermediate data
• DNN-layers Parallelization: Effective multi-core utilization for elevated performance & optimized core-to-core data transfer
• Aggressive Quantization: Minimization of quantization loss through mixed-precision computation

• Object detection and tracking
• Face detection and identification
• Human pose detection, gesture recognition
• Vision-based defect inspection
• Natural language interface

Documentation

• NPU HW integration guide
• NPU SW toolkit guide
• Linux device driver & API manual
• Technical reference manual

NPU HW

• RTL (Verilog)
• Example testbench
• Synthesis constraints

NPU SW

• Network compiler toolkit
• Linux device driver