Low-Power ASIC Clusters

Application and Function Overview: High-Density Computing

Unlike high-performance AI chips, the advantage of low-power ASICs lies in their high energy efficiency (P/W) and low cost. They achieve performance through the integration of hundreds or even thousands of dedicated computing cores, excelling in scenarios where tasks can be highly parallelized. Typically, several computing boards are placed in a cabinet, with each board housing hundreds of ASICs.

Application Areas and Architectural Features

Application Area Core Tasks Structural Features
Data Encryption/Hash Computation Highly repetitive integer hash calculations Array-based ASIC cores, low power consumption, high density
Edge AI Inference Sensor data preprocessing, small model inference NPU array, aiming for extremely low latency inference with INT8/INT4

Low-Power ASIC Array System Architecture Diagram

Low-Power ASIC Clusters

Key Components and Functional Differences

Unlike high-end general-purpose computing cards, low-power ASIC cards emphasize multi-channel power distribution and high-density interconnects at the board level.

Function Category Core Components Overview Function
Computing Core ASIC array chip (NPU/Hash Core) A large number of computing cores executing a single/simplified instruction set, pursuing extreme energy efficiency.
High-Speed Interconnect PCIe Bridge / Local Switch Responsible for efficiently distributing host tasks to the hundreds of onboard ASIC chips.
Power Management Multi-channel DC-DC VRM modules Responsible for converting input voltage into multiple ultra-low voltage, high current outputs, evenly distributed across the entire ASIC array.
Local Storage Small capacity DRAM / SRAM Used only for storing the inputs and outputs of current computing tasks, with bandwidth requirements far lower than HBM.

Bill of Materials (BOM)

Component Name Common Models Alternative References
ASIC Array Chip BM1397 (Low-Power Hash Core) or Custom NPU Cambricon SiYuan Edge Series, Bitmain BM Series
PCIe Bridge/Switch PLX PEX Series or Marvell Lanqi Technology PCIe Retimer / Jingjia Micro JM Series
Small Capacity DDR4 / SRAM Micron DDR4/3L (Low-Power Version) Changxin Memory (CXMT) DDR3/4
Multi-Phase PWM Controller Renesas ISL Series or TI TPS Series Southchip SC30xx Series
Onboard MCU (Low-Power) STM32 Series (F1/F4) GigaDevice (GD) GD32 Series / Huada Semiconductor HC32

Related posts

  1. Accelerated Return of Overseas Talent: Intel Chip Architect Su Fei Returns to Tsinghua University After 20 Years in the U.S.
  2. Python Statement Structures
  3. Introduction to the Automation Operations Tool Ansible Role
  4. Embedded Innovation Project Workshop: Become a Relentless Traveler
  5. Day 4 of Python: Mastering 3 New Containers and Advanced Loops
  6. How to Play Windows Games on Linux in 2002
  7. ASIC Chips: Market Leaders Seize Opportunities
  8. Low Power Design: The Energy-Saving Code of Electronic Products
  9. Top 10 Low-End Leaders in ASIC Chips: Don’t Wait Another Ten Years!
  10. Broadcom Secures $10 Billion Order, ASIC Chips Quietly Disrupting AI Computing Landscape!
  11. Google TPU Chip (A Type of ASIC Chip) Concept Stocks
  12. Global AI-MCU Market Size and Future Growth Potential
  13. MCU Micro Classroom | CKS32K148 LPI2C
  14. A Bioinspired Lipopeptide Surmounts Therapeutic Dilemmas of Gram-Positive and Gram-Negative Polymicrobial Infection
  15. Using Combinational Logic or Sequential Logic for Output Signals in FPGA
  16. Low Power Design Techniques
  17. Microcontroller Low Power Design Techniques: Secrets to Energy Saving and Environmental Protection
  18. Learning C Language from Scratch: Chapter 7: Functions and Scope
  19. [RTOS] The Night Before Scheduling: Analysis of Interrupt Failure in FreeRTOS

Leave a Comment