Sparse data structures like hash tables, trees, and compressed tensors, are widely used in many important applications, including databases, graph analytics, genomics, scientific computing, sparse linear algebra, and deep learning. However, operations on these data structures often suffer from efficiency issues due to data dependent control operations, unable to be optimized through branch prediction, and long-latency memory accesses, such as those found while traversing trees. Terminus is a simple programmable accelerator that contains a spatial compute fabric and delegates tasks to several processing elements that in tandem exploit concurrency to introduce speed up over reads and writes. This results in exceptional speed ups across several applications.

I am pursuing this project as a means to develop my skills and knowledge as an engineer and to eventually make an educated choice as to pursue either academia or industry in the future.