Near Memory Processing Architecture for Deep Learning Models

Design Doc

Specs

DNN Target Specs

Data-Stream Specs

• Input Size
• Input Shape
• Input Rate

Hardware Specs

• Processing Engine(PE) Specs
  • Speed
  • Energy Consumption
• Outer SRAM Buffer
  • Speed
  • Energy Consumption
• 3D HMC RAM Spec
  • Speed
  • Energy Consumption
• 2D DRAM Spec
  • Speed
  • Energy Consumption

Scheduling and Allocation Algo

• DL Model IR

  • Convert the DL model into universal representation with respect to model and input size

• Allocation Algo

• Pipeline Scheduling Algo

Simulator