Personal tools

Difference between revisions of "Analog Compute-in-Memory Accelerator Interface and Integration"

From iis-projects

Jump to: navigation, search
Line 46: Line 46:
 
[[Category:Master Thesis]]
 
[[Category:Master Thesis]]
 
[[Category:Hot]]
 
[[Category:Hot]]
 +
[[Category:Heterogeneous Acceleration Systems]]
 +
  
 
===Practical Details===
 
===Practical Details===

Revision as of 16:21, 18 October 2021

FeFET ACiM.png

Description

The continuous development and use of computation- and memory-intensive algorithms, such as Deep Neural Networks (DNNs) are currently being limited by the substantial energy and latency needs of off-chip memories. To tackle this issue, researchers have started investigating on structures base on non-volatile devices, that can both perform logic and arithmetic operations, and function as memories, thus starting the in-memory computing (IMC) paradigm.

One of these most novel and promising devices are the ferroelectric FETs (FeFETs), which exploit a thin ferroelectric layer between gate and channel to store data [1]. These devices can be exploited as the basic cell of analog compute-in-memory (AciM) accelerator for MAC operation that can significantly outperform the current state-of-the-art in terms of power efficiency (TOPS/W), retention time, and area scalability [2].

The target for this project is to design the interface for the ACiM accelerator and integrate the accelerator into a modern microcontroller system. In this project the student will: 1. Work in close contact with industry partners for the development of a FeFET IMC crossbar (XBAR) array 2. Design interface for ACiM accelerator 3. Integrate the accelerator with MCU 4. Verify functionality of the system

Status: Available

Looking for master or semester thesis students
Supervisor: Giorgio Cristiano, Jiawei Liao

Prerequisites

  • Worked with at least one RTL language in the past (SystemVerilog or Verilog)
  • Prior knowledge of hardware design and computer architecture
  • VLSI I
  • Knowledge of analog circuit is a plus (e.g. AIC)

Character

  • 20% Literature review
  • 20% Architecture Design
  • 30% RTL implementation
  • 30% Verification

Professor

Prof. Taekwang Jang

Reference

[1] T. Soliman et al., "A Ferroelectric FET Based In-memory Architecture for Multi-Precision Neural Networks," 2020 IEEE 33rd International System-on-Chip Conference (SOCC), 2020, pp. 96-101, doi: 10.1109/SOCC49529.2020.9524750.


[2] T. Soliman et al., "Ultra-Low Power Flexible Precision FeFET Based Analog In-Memory Computing," 2020 IEEE International Electron Devices Meeting (IEDM), 2020, pp. 29.2.1-29.2.4, doi: 10.1109/IEDM13553.2020.9372124.

↑ top


Practical Details