NPU IP for AI Vision and AI Voice

The VIP9000 family offers programmable, scalable and extendable solutions for markets that demand real time and low power AI devices. VIP9000 Series’ patented Neural Network engine and Tensor Processing Fabric deliver superb neural network inference performance with industry-leading power efficiency (TOPS/W) and area efficiency (mm2/W). The VIP9000’s scalable architecture, ranging from 0.5TOPS to 20TOPS, enables AI capability for a wide range of applications, from wearable and IoT devices, IP Cam, surveillance cameras, smart home & appliances, mobile phones and laptops to automotive (ADAS, autonomous driving) and edge servers. In addition to neural network acceleration, VIP9000 Series are equipped with Parallel Processing Units (PPUs), which provide full programmability along with conformance to OpenCL 3.0 and OpenVX 1.2.  

VIP9000 Series IP supports all popular deep learning frameworks (TensorFlow, TensorFlow Lite, PyTorch, Caffe, DarkNet, ONNX, Keras, etc.) and natively accelerates neural network models through optimization techniques such as quantization, pruning, and model compression. AI applications can easily port to VIP9000 platforms through offline conversion by Vivante’s ACUITYTM Tools SDK or through run-time interpretation with Android NN, NNAPI Delegate, ARMNN, or ONNX Runtime.

• Programmable Engines (PPU)
  • 128-bit vector processing unit (shader + ext)
  • OpenCL 3.0 shader instruction set
  • Enhanced vision instruction set (EVIS)
  • INT 8/16/32b, Float 16/32b
• Tensor Processing Fabric
  • Non-convolution layers
  • Multi-lane processing for data shuffling, normalization, pooling/unpooling, LUT, etc.
  • Network pruning support, zero skipping, compression
  • On-chip SRAM for DDR BW saving
  • Accepts INT 8/16b and Float16 (Float16 internal)
• Unified Programming Model
  • OpenCL, OpenVX, OpenVX-NN Extensions
  • Parallel processing between PPU and NN HW accelerators with priority configuration
  • Supports popular vision and deep learning frameworks: OpenCV, Caffe, TensorFlow, ensorFlowLite, ONNX, PyTorch, Darknet, Keras
• SW & Tools
  • ACUITY Tools: End-to-end Neural Network development tools
  • Eclipse-based IDE for coding/debugging/Profiling
  • NNRT: Runtime framework supporting a droid NN, NNAPI Delegate, ONNX Runtime and ARMNN.
• Scalability
  • Number of PPU and NN cores can be configured independently
  • Same OpenVX/OpenCL code runs on all processor variants; scalable performance
• Extendibility
  • VIP-Connect^TM: HW and SW I/F protocols to plug in customer HW accelerators and expose functionality via CL/VX custom kernels
  • Reconfigurable EVIS allows user to define own instructions
  • Easy integration with other VSI IPs