Ultra-precise platform for scientific calculation (ArTeCS)

The research activity of the Group of Architecture and Technology of Computing Systems (ArTeCS) of the Complutense University of Madrid (UCM) is focused on the conception and construction of digital information processing systems, and its efficient application regarding performance, energy consumption, and cost. Within this broad area, the group pays special attention to high-performance computing, processor & memory hierarchy design and embedded systems. The research activity of the ArTeCS group covers a wide range of subjects related with high performance computing, computer architecture and system design. Besides, the ArTeCS group is one of the groups in UCM graded as "Excellent" by an evaluation commissioned by the State Research Agency. Two of the most relevant researching lines of the group is the use of Specialized Arithmetic and Hardware Acceleration, that are the basis of the ASIMOV project, starting on Sept. 1st, 2022 and whose PI is Dr. Alberto del Barrio (grant PID2021-123041OB-I00 funded by MCIN/AEI/ 10.13039/501100011033 and by “ERDF A way of making Europe”). In relation to the project, recently we have released PERCIVAL, that is our posit-enabled RISC-V core integrating the xPosit extension (https://github.com/artecs-group/PERCIVAL), and Flo-Posit, a repository with posit arithmetic units (https://github.com/RaulMurillo/Flo-Posit) employed in PERCIVAL. The latter is the main outcome of the PARNASO project, awarded by the 2020 Leonardo Grants program of the BBVA Foundation and whose PI has been Dr. Alberto del Barrio.

Data-intensive applications have permeated many areas of computing due to the increasing demand for larger resolutions (e.g., molecular dynamics, weather forecasting, quantum computing simulations). All these applications require an extremely high precision to produce accurate results, so they typically leverage double (64-bits) or even quadruple precision (128-bits) of the IEEE-754 standard, with the consequent impact on power. Recently, new arithmetic formats as the posits, aka universal numbers version 3, have appeared for representing real numbers. Posits were introduced by John L. Gustafson in 2017 to mitigate the problems inherent in floating point, defined by the IEEE-754 standard. This format promises to achieve dynamic ranges similar to the floating point, reducing the bitwidth by half, so its impact on consumption, memory latency, operations execution time and more on any hardware system can be enormous. Besides, posits leverage a large accumulator called quire that avoid losing precision in data-intensive calculations, so they are very well suited for scientific calculations. However, due to the novelty of the posits, they are not standard and in general they need to be emulated through software libraries, which implies a huge loss of performance as soon as the application has a certain complexity. In the last two years several RISC-V-based processors have appeared in literature to integrate posit calculations, PERCIVAL being one of the most complete. Nevertheless, there are open challenges such as: improving programmability capabilities going beyond assembly, lack of high-level compiling support or the deployment of parallel architectures. The objective of this project is to mitigate the afforementioned problems by constructing a SW/HW environment able to provide an ultra-accurate platform to perform scientific calculations, as well as a friendly programming environment to make the programming task easier.

A motivation letter identifying research synergies (max 1 page); your CV (including a list of publications) limited to a maximum of 4 pages; short summary of your proposed project idea (max 1 page)