Center for Research in Intelligent Storage and Processing in Memory

CRISP concluded in September 2023. This page serves as an archive reference point. If you have questions about CRISP work, please email Kevin Skadron at [email protected]. 

About CRISP

Computer systems are increasingly bottlenecked by data movement costs and are no longer able to keep up with the “big data” explosion. This trend is compounded by the fact that data in memory is stored without any consideration of data content, usage patterns, or how data are grouped into higher-level objects. Many algorithms also exhibit poor temporal locality (because they only touch data items once) and poor spatial locality (because they exhibit irregular access patterns across vast data sizes), making caches inefficient.

Solving these challenges and enabling the next generation of data-intensive applications requires computing to be embedded in and around the data, creating intelligent memory and storage (IMS) architectures that do as much of the computing as possible as close to the bits as possible. The closer computation is to the memory cells, the closer it comes to the raw bandwidth available in the storage arrays. However, such IMS capabilities will require reconstructing the entire system stack, with new architectural and operating-system abstractions, new memory semantics, and new techniques for compiling and optimization, and dynamic yet efficient system software. At the same time, we want to achieve high programmer productivity and code portability across diverse, heterogeneous architectures. We must protect the user from this hardware and software ferment by providing a highly intuitive programming model that allows programmers to focus on what they want to do, instead of how, while nevertheless operating as close to the raw performance of the data arrays as possible.

The CRISP grand challenge is to significantly lower the effort barrier for every day programmers to achieve highly portable, “bare-metal,” and understandable performance across a wide range of heterogeneous, IMS architectures. This will democratize high-performance, heterogeneous, data-intensive computing, to enhance productivity of the IT workforce and enable an improved software ecosystem that opens new markets for computer systems.