Specialized Video Processing NPU IP

Overview

The WAVE-N is a custom NPU designed to work for imaging application at high-quality and high-resolution in edge devices. Compared to typical generic NPU, it provides extreme computation power in 16bit floating point format for high performance together with competitive enough area and power efficiency.

Target Performance

C&M’s imaging solutions powered by WAVE-N, as of 2Q 2025 C&M currently offers three types of neural networks:

  • SR | Scale X2 : 1920x1080 → 3840x2160 60fps @500MHz
  • NR |  RGB / Bayer : 1920x1080 60fps @500MHz
  • NR+Demosaic | 8M(4096x2048) sensor 30fps @500MHz

Key Features

Fully Programmable Processing Core

  • Instruction level coding with C&M proprietary ISA 
  • (Instruction Set Architecture)

16bit Floating Point Arithmetic Unit

  • High-quality image output (8-bit, 10-bit)
  • No risk in fixed-point arithmetic error
  • No risk in quantization aware network training

 No DRAM Bandwidth 

  • No access to external DRAM in internal 
  • Processing to handle map data
  • Line by Line processing

 Extreme Efficiency for Modern CNN Computation

  •  Up to 90% MAC utilization
  • 1024 16bit floating point MAC (2TFLOPS)

Benefits

  • Optimized for Real-time
  • FP16 NPU with Area efficiency
  • Minimum DRAM Bandwidth
  • Fully Programmable Core

Block Diagram

Specialized Video Processing NPU IP Block Diagram

Applications

  • Mobile,
  • Wearable,
  • Drone,
  • IP Camera,
  • Digital Camera,
  • TV,
  • Set-top box,
  • Automotive

Technical Specifications

Short description
Specialized Video Processing NPU IP
Vendor
Vendor Name
×
Semiconductor IP