Journal Highly available systems №3 for 2018 г.
Article in number:
Configurable special VLSI – real basis for future exascale supercomputers (foreign and domestic experience)
Type of article: scientific article
DOI: 10.18127/j20729472-201803-04
UDC: 004
Keywords: high-end (HEC) supercomputers supercomputer electronics components blockchain deep learning configurable problem-oriented special VLSI programmable unit reconfigurable unit energy efficiency
Authors:

L.K. Eisymont – Ph.D.(Phys.-Math.), The Federal Register of Scientific and Technical Experts  of the Russia Ministry of Education and Science (Moscow) E-mail: verger-lk@yandex.ru

Abstract:

The direction of creation of problem-oriented specialized VLSI (POS-VLSI) as a variant of asymmetric replacement of the foreign element-component base (ECB) in domestic supercomputers and cloud servers of the highest performance range is considered. Configurable POS-VLSI, in contrast to VLSI with direct hardware implementation of algorithms (non-configurable), have the ability to implement them by programming or reconfiguring their blocks. Some POS-VLSI have a weak or strong specialization. Strong specialization is most important, but it is difficult to do without waste of productivity and energy efficiency. This is a serious scientific and technical problem, the emerging options for its solution require careful study. In the world practice, the development of PIC-VLSI is also underway, Chinese developers have made special progress. In the world practice, the development of this type of POS-VLSI is also under way, Chinese developers have made special progress. Judging by the results achieved, this becomes a strategy for ECB development under threatening restrictions because of the approaching end of the development of CMOS technologies and the way ahead of the rival countries in the field of high technologies. The article systematizes and analyzes achievements in this field, compares it with domestic developments.

Pages: 18-27
References
  1. Eisymont L.K. Gibridnaya strategiya razvitiya ehlementnoj bazy // Otkrytye sistemy. 2017. № 2. URL = https://www.osp.ru/os/2017/ 02/13052216/.
  2. Sadsiven S.K., Tompto B.W., Kalla R., Starke W.J. IBM Power 9 Processor Architecture. IEEE Micro. March/April 2007. P. 40−51.
  3. Kogge P.M., Brockman J.B., Harper D.T., Smith B., Callahan C.D. Computer systems with lightweight multi-threaded architectures. United States Patent Application Publication. Pub. No.: US 2007/0198785 A1. Pub. Date: Aug. 23. 2007. 34 p.
  4. Schor D. ORNL’s 200-petaFLOPS Summit Supercomputer Has Arrived, To Become World’s Fastest. URL = https://fuse.wikichip.org/ news/1351/ornls-200-petaflops-summit-supercomputer-has-arrived-to-become-worlds-fastest/.
  5. Durant L. et al. Inside Volta: The World’s Most Advanced Data Center GPU. 10 May 2017. URL = https://devblogs.nvidia.com/parallelforall/inside-volta/.
  6. Sluckin A.I., Eisymont L.K. Rossijskij superkomp'yuter s global'no adresuemoj pamyat'yu // Otkrytye sistemy. 2007. № 9. S. 42−51. URL = http://www.osp.ru/os/2007/09/4569294/.
  7. Zhabin I., Makagon D., Simonov A., Syromyatnikov E., Frolov A., Sherbak A. Kristall dlya «Angary» // Superkomp'yutery. 2013. № 4(16). S. 46−49.
  8. Eisymont L.K., Semenov A.S., Sokolov A.A., Frolov A.S., Shvorin A.B. Modelirovanie rossijskogo superkomp'yutera «Angara» na superkomp'yutere. Superkomp'yuternye tekhnologii v nauke, obrazovanii i promyshlennosti / Pod red. akad. V.A. Sadovnichego, akad. G.I. Savina, chl.-korr. RAN Vl.V. Voevodina. M.: Izd-vo Moskovskogo universiteta. 2009. 5 s. URL = http://hpc-russia.ru/book1/18.pdf.
  9. Khazraee M., Gutierrez L.V., Magaki I., Tailor M.B. Specializing a Planet’s Computation: ASIC Clouds. IEEE Micro. May/June 2017. P. 62−69.
  10. Wei L., Xiaoyang Z., Zibin D., Longmei N., Tao C., Chao M. A High Energy-Efficient Reconfigurable VLIW Symmetric Cryptographic Processor with Loop Buffer Structure and Chain Processing Mechanism // Chinese Journal of Electronics. Nov. 2017. V. 26. № 6. P. 1161−1167.
  11. Novatsky T. et al. Pashing the Limits of Accelerator Efficiency While Retaining Programmability // IEEE High performance computer architecture conference. 2016. 13 p. URL = http://pages.cs.wisc.edu/~vinay/pubs/hpca16-lssd.pdf.
  12. Govindaraju V. et al. Dynamically Specialized Datapaths for Energy Efficient Computing // 17th IEEE International Symposium on High Performance Computer Architecture. 2011. 12 p. URL = http://research.cs.wisc.edu/vertical/papers/2011/hpca11-dyser.pdf.
  13. Sayliar G. G., Chiou D. Cryptoraptor: High Throughput Reconfigurable Cryptographic Processor // IСCAD 2014. 8 p. URL = https://pdfs.semanticscholar.org/6d0e/a468a737ef666d6a9da0ee1c2f1cff3f43b3.pdf.
  14. Dally W., Balford J. et al. An Energy-Efficient Processor Architecture // IEEE Computer Architecture Letters. 2008. V. 7. № 1. 4 p. URL = http://cva.stanford.edu/projects/elm/pubs/cal_Jan_2008.pdf.
  15. Ajavi T. et al. Experiences Using the RISC-V Ecosystems to Design an Accelerator-Centric SoC in TSMC 16 nm // First Workshop on Computer Architecture Research with RISC-V. August 2017. 6 p.
  16. Elizarov S.G., Luk'yanchenko G.A., Markov D.S., Monahov A.M., Roganov V.A., Sizov A.D. Programmiruemye na yazykah vysokogo urovnya specializirovannye SBIS dlya zadach informacionnoj bezopasnosti s predel'noj ehnergoehffektivnost'yu // Sb. dokladov VI Mezhdunar. nauchno-praktich. konf. «Upravlenie informacionnoj bezopasnost'yu v sovremennom obshchestve». 5−7 iyunya 2018. Moskva.
  17. Ejsymont A.L., Chernikov A.V., Kosorukov D.E., Nasonov I.I., Komlev A.A. Geterogennaya mnogoprocessornaya sistema na kristalle s proizvoditel'nost'yu 512 Gflops // Tekst doklada. Mikroehlektronika-2017. Alushta. 11 s. URL = https://module.ru/upload/images/ 1507808903Гетерогенная многопроцессорная СнК с производительностью 512 Gflops.pdf
  18. Chernikov V.M., Viksne P.E. Perspektivy povysheniya harakteristik i rasshireniya oblastej primeneniya transteraflopsnyh SBIS semejstva NeroMatrix // Sb. dokladov VI Mezhdunar. nauchno-praktich. konf. «Upravlenie informacionnoj bezopasnost'yu v sovremennom obshchestve». 5−7 iyunya 2018. Moskva.
  19. Broad Agency Announcement Hierarchical Identify Verify Exploit (HIVE) Microsystems Technology Office DARPA-BAA-16-52. August 2. 2016. 50 p.
  20. Song W.S., Kepner J., Gleyzer V., Nguyen H.T., Kramer J.I. Novel Graph Processor Architecture // Lincoln Laboratory Journal. 2013. V. 20. № 1. P. 92−104.
  21. Broad Agency Announcement Circuit Realization At Faster Timescales (CRAFT) Microsystems Technology Office DARPA-BAA-15-55. August 17. 2015. 46 p.
Date of receipt: 3 августа 2018 г.