Scalable, Configurable Neural Network Accelerator Based on RISC V Core

on YouTube ↗09m

" summary

LG Neural Engine (LNE) is a configurable NN accelerator built around a RISC-V Rocket core (4‑stage RV32I, M/C optional) plus a tiled NPU, silicon proven on TSMC 28nm HPC+. Architecture scales to 128 cores (tiles of NPUs), each tile hosting on‑chip man‑blobs and a shared TSM, with a data‑movement engine (DMA) supporting 2D/3D layouts, multicast/broadcast and AXI host interfaces. NPU datapath runs 16‑bit (8x8) and 8‑bit (16x16) MAC arrays, offers binarized weights for higher FPS, and targets 1 GHz operation.

AI-written summary · about

" chapters (12)

1. 0:00Introduction
2. 0:34LG Neural Engine
3. 0:52LNE's Features
4. 2:00LNE's Architecture
5. 3:47Neural Processing Unit (NPU)
6. 4:32Common iCache/Data Memory
7. 4:55MBLOBs and TSM
8. 5:28Data Movement Engine (DME)
9. 6:05Host Access
10. 6:28Ring Bus
11. 7:28LNE's Software Flow
12. 7:49Demo

" related talks

• 09mA Tightly Coupled Light Weight Neural Network Processing Units With RISC V CoreLei Zhang · 7th RISC-V Workshop 2017
• 13mKeynote: Taking RISC-V Intelligence EverywhereCharlie Su · RISC-V Summit North America 2023
• 22mKeynote: RISC-V at NVIDIA: One Architecture, Dozens of Applications, Billions of ProcessorsFrans Sijstermans · RISC-V Summit North America 2024
• 19mRISC-V: Changing the Way AI/ML Accelerators and Computing Infrastructure Are BuiltDavid Chen · RISC-V Summit North America 2024
• 53mBuilding an AI Accelerator with High-Level SynthesisWerner Bachhuber · RISC-V Summit Europe 2026 Developer Workshops
• 24mIndustrial Strength High Performance RISC V Processors For Energy Efficient ComputingDavid Ditzel · 7th RISC-V Workshop 2017
• 24mReal World Results using Thousands of RISC-V Cores for AI and BeyondDave Ditzel · RISC-V Summit 2022
• 10mBoosting AI on Semidynamics RISC-V Cores with Custom Tensor InstructionsRoger Espasa · RISC-V at embedded world 2024