Details
Presenter(s)
![Abdullah Mansoor Headshot](https://confcats-catavault.s3.amazonaws.com/CATAVault/ieeecass/master/files/styles/cc_user_photo/s3/user-pictures/17941_1.jpg?h=e234a08a&itok=-cXm4i-k)
Display Name
Abdullah Mansoor
- Affiliation
-
AffiliationPortland State University
- Country
-
CountryUnited States
Abstract
The proposed Reinforcement Learning (RL) and Simulated Annealing (SA) based placement (RS3DPlace) algorithm for Monolithic 3D (M3D) ICs is the first Machine Learning approach for M3D. It uses an approximate method for state representation with reduced memory complexity compared to previously published works restricted to 2D only. Our current implementation is for the gate-level M3D design style, but it can be extended to other M3D styles. RS3DPlace solves placement of 896 variables compared to a maximum size of 625 variables in the previously published work. RS3DPlace results show an average 16% improvement in overall objective function in comparison to SA.