ChipMi

Using a novel spatial graph embedding method, we propose environmentally focused chip design optimizations that will mitigate waste and thermal issues common in current manufacturing processes.

Comment

Hardware is the backbone of technological information. The silicon chips integrated throughout our society come at a high demand, but with severe environmental costs such as energy and water consumption. To combat this issue, we propose a novel design system for optimizing the construction of silicon chips, focused on environmentally conscious manufacturing processes and thermal management. Our approach uses metric2vec, a novel spatial graph traversal method, which can identify patterns of co-occurrence among nodes of a graph through the lens of specific metrics; the choice of lens being the maximum flow between all pairs of nodes. This in turn enables us to analyze congestion and disjoint paths in the graph. We specifically choose the maximum flow metric for analyzing the spatial aspects of chip development. We used Modal Lab’s cloud computing technologies to host our website, in which users can input their desired chip specifications to yield as output an embedding of their graph such that points close together in the embedding represent nodes in the original graph which have high flow of traffic. A user can define the graph of the chip they would like designed in NetworkX edgelist format, to which our website will parse and output a graphical interpretation, which we believe to provide spatial awareness of how the chip components should be placed. Our approach focuses on user accessibility in chip design, while also providing novel hardware approaches for tackling the environmental aspects of the technological sector. Challenges we faced included algorithm design and implementation, as this is the first time the graph-based method has been used. Three of the members were first time hackers, so learning the basics of frontend design with CSS and HTML paired with backend implementation using Modal Lab's cloud computing and Flask was a learning extravaganza. We are proud of the potential impact on chip design, and are excited for the other applications of the graph-based method to differential spatial problems under different metrics. Aside from immediate software design, the team gained experience in collaboration and the implementation of mathematical abstractions. We plan to expand our algorithms capabilities to other problems, and create a 5c club focused on solving these issues.

Built With

Share this project:

Updates