DC25_PAPER_TRACK07_GeneticAlgorithmDrivenIBISAMI_Auge.pdf

1、 Information Classification:General Genetic Algorithm-Driven IBIS-AMI Optimization for Robust 200G SerDes Design Adrien Auge,Alphawave Semi Tripp Worrell,MathWorks Walter Katz,MathWorks Richard Allred,MathWorks Eric Brock,MathWorks Information Classification:General Abstract Copper connectivity is i

2、ncreasingly becoming a bottleneck in the rapidly evolving compute landscape that supports artificial intelligence(AI).The design of 200G SerDes faces aggressive timelines,necessitating a robust pre-silicon simulation methodology.A significant performance limiter in SerDes technology is the adaptatio

3、n routine that determines the optimal equalization(EQ)for a given channel.The routines effectiveness,whether in laboratory settings or simulations,heavily depends on the underlying optimization algorithms robustness.Due to runtime constraints,traditional simulation models often fail to represent har

4、dware precisely when searching for optimal equalization recipe instead focusing on achieving the required simulation throughput these types of electrical simulations demand by adding a certain level of abstraction.Even with modeling simplification,proper adaptation of the model equalization settings

5、 necessitates a considerable amount of the total computation to be put on this optimization task.This paper proposes enhancements to existing modeling techniques to enable a faster and more accurate representation of SerDes,thus narrowing the gap between model abstraction and actual silicon/firmware

6、 performance.By adopting proven optimization strategies,such as Genetic Algorithms,over traditional brute-force methods for driving adaptation,we aim to improve the efficiency at arriving at an optimal equalizer recipe for a given channel.Along with efficiency improvement in optimizing the SerDes re