(6th-February-2020)
Figure 3 MDGRAPE-4A system
In the development of the MD dedicated computer, we independently designed an LSI that integrated memory (GM), general-purpose processor (GP), particle interaction pipeline (PP), and network interface (NIF) on a chip. As shown in the four rows on the left, 64 boards on which 8 LSIs are mounted are connected by optical connection and FPGA system, and one MD calculation is performed at high speed as a whole.
Future expectations
The in silico drug discovery technology has the great advantage that it can be implemented using the structural formulas of candidate molecules, and the screening can be performed on almost unlimited number of chemical structural formulas. By performing a long-term simulation with MDGRAPE-4A, it is possible to search for structural changes when a candidate molecule and a protein actually bind, and achieve more accurate predictions. In addition, this calculation method is promising in the development of molecules that control not only the “form” but also the “motion” of proteins, and can be expected to greatly expand the potential of drug discovery. In the future, we would like to share this with industry and academia in a wide range of fields other than drug discovery.
In the area of computer development, the role of dedicated computing systems is likely to become increasingly important as Moore's Law [13], which has supported semiconductor performance improvements, has come to an end. Only the DE Shaw Research Institute in the United States [14] and the RIKEN research team are currently working on “deep-level integration” of both hardware and software in MD calculations. In addition to further acceleration of MD calculation, development of large-scale system by combining dedicated circuit and general-purpose circuit, such as application to artificial intelligence.
Commenti