Table 4.2: Comparison in terms of average error between conventional and stochastic approaches in case of injecting different ratios of soft errors
Chapter 5
Conclusions and Perspectives
This master thesis presents the principal problems with respect to Stochastic computing. Such kind of computing is generally an alternative choice to conventional computing methods in which numbers are interpreted as probabilities and represented by bit sequence that can be implemented by very simple circuits. The main advantages of SC are : very low–cost circuit, highly fault–tolerant to manufacturing process variations and soft errors. Nervertheless, it also faces the long computation time and accuracy degradation due to its stochastic features. Three principal problems are considered in this work:
Concerning the first issue, there are the trade–off between the accuracy and hardware cost in case of in- vestigating LFSR–based PRNGs: classical LFSR approach, multiple LFSRs approach and Leap–Ahead LFSR approach. The third one give the generation of SNs in parallel, and consume less hardware than others. Meanwhile, SNGs relying on CA–based PRNG with the good features of CA provide an out- standing quality compared the previous approaches, however, this technique requires more complexity in hardware configuration. Furthermore, there are the methods that using a number of PRNG and basic logic elements to generate SNs with a considerable accuracy. Besides, FSM has recently investi- gated in SC, hence, a synthesis approach in condition of random source limitation was also analysed in our work.
While SNGs are the most important problem in SC, stochastic operators equally play an essential role in this system. First, the stochastic basic elements which was early proposed such as AND gate, XNOR gate, MUX which consume less hardware in carrying out the unsigned multiplication, signed multiplication, scaled adder, respectively in conventional computing. Furthermore, FSM–based stochastic elements have proved their outstanding characteristics in carrying out the complexe functions as tanh, absolute, exponential exponentiation on absolute value. Besides, based on the previous works, we also develope the triple-length LFSR based Stochastic scale adder which can be give a new way to implement the scale addition in SC.
Our work investigated the above stochastic operators in image processing applications. The first one, edge detection was accomplished by using two operators: Roberts and Prewitt. According to the received results, it can be shown that SC with its simplicity in hardware architecture can give the output as good as that of conventional computing. For the second operator, noise reduction based on median filter equally yields the results of high quality. The simulation on different lengths of this operator is also executed, and as expected, the accuracy and precision increase steadily with stochastic sequence length. This part ended by evaluating the faul–tolerant capacity of SC in comparison with traditional approach. Our experiments presented a slow change in stochastic design to manufacturing process variations and soft errors, while conventional computing is very sensitive with such changes. Although due to the requirement of a large number of bits to represent, it is evident to see that SC consumes more energy compared traditional approaches, it might be the amenable choice when the transient circuit variations are present.
SC, an unconventional computing which exploits the probabilistic aspect of traditional technologies, brings a new way which is effective in some applications where uncertainty is considered. It can be seen as a method of simple hardware implementation and high fault tolerance. However, the accuracy and processing time are the main problems which are still poorly understood. Therefore, the following issues might be taken into account in the future:
been on SC designs with only few stages, hence the complexity may be still not considerable. Future works must deal with more stages, and a method to evaluate the accuracy of general SC systems is therefore an essential requirement.
To sum up, SC has the outstanding features which can adapt to the current technology trends where the uncertainty in circuit behavior becomes more interesting than ever. It will definitely lead to long–term study in the future.
Bibliography
[1] http://en.wikipedia.org/wiki/Uniform_distribution_(continuous).
[2]A. Alaghi and J.P Hayes. A spectral transform approach to stochastic circuits. In IEEE 30th International Conference on Computer Design (ICCD), pages 315–321. IEEE, 2012.
[3]A Alaghi, Cheng Li, and J.P. Hayes. Stochastic circuits for real-time image-processing applica- tions. In Design Automation Conference (DAC), 2013 50th ACM / EDAC / IEEE, pages 1–6, May 2013.
[4]B.D. Brown and H.C. Card. Stochastic neural computation. i. computational elements. Comput- ers, IEEE Transactions on, 50(9):891–905, Sep 2001.
[5]B.D. Brown and H.C. Card. Stochastic neural computation. ii. soft competitive learning. Com- puters, IEEE Transactions on, 50(9):906–920, Sep 2001.
[6]Yun-Nan Chang and K.K. Parhi. Architectures for digital filters using stochastic computing. In Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on, pages 2697–2701, May 2013.
[7]J.M. Comer, J.C. Cerda, C.D. Martinez, and D.H.K. Hoe. Random number generators using cellular automata implemented on fpgas. In System Theory (SSST), 2012 44th Southeastern Symposium on, pages 67–72, March 2012.
[8]B.R Gaines. Stochastic computing. In Proc. AFIPS Spring Joint Computer Conf., pages 149–156.
ACM, 1967.
[9]B.R. Gaines. Stochastic computing systems. In JuliusT. Tou, editor, Advances in Information Systems Science, Advances in Information Systems Science, pages 37–172. Springer US, 1969.
[10]Rafael C. Gonzalez and Richard E. Woods. Digital Image Processing (3rd Edition). Prentice-Hall, Inc., Upper Saddle River, NJ, USA, 2006.
[11]Sheng-Uei Guan and S.K. Tan. Pseudorandom number generation with self-programmable cellular automata. Computer-Aided Design of Integrated Circuits and Systems, IEEE Transactions on, 23(7):1095–1101, July 2004.
[12]P. K. Gupta and R. Kumaresan. Binary Multiplication with PN Sequences. Acoustics, Speech and Signal Processing, IEEE Transactions on, 36(4):603–606, Apr 1988.
[13]Je-Hoon Lee and Seong Kun Kim. Segmented leap-ahead lfsr architecture for uniform random number generator. In International Journal of Software Engineering and Its Application, pages 233–242, 2013.
[14]Peng Li and D.J. Lilja. Using stochastic computing to implement digital image processing al- gorithms. In Computer Design (ICCD), 2011 IEEE 29th International Conference on, pages 154–161, Oct 2011.
[15]Peng Li, D.J. Lilja, W. Qian, M.D. Riedel, and K. Bazargan. Logical computation on stochastic bit streams with linear finite-state machines. Computers, IEEE Transactions on, 63(6):1474–1486, June 2014.
[16]Peng Li, D.J. Lilja, Weikang Qian, K. Bazargan, and M.D. Riedel. Computation on stochastic bit streams digital image processing case studies. Very Large Scale Integration (VLSI) Systems, IEEE Transactions on, 22(3):449–462, March 2014.
[17]Peng Li, Weikang Qian, M.D. Riedel, K. Bazargan, and D.J. Lilja. The synthesis of linear finite state machine-based stochastic computational elements. In Design Automation Conference (ASP- DAC), 2012 17th Asia and South Pacific, pages 757–762, Jan 2012.
[18]Xin Li, Weikang Qian, Marc D. Riedel, Kia Bazargan, and David J. Lilja. A reconfigurable stochastic architecture for highly reliable computing. In Proceedings of the 19th ACM Great Lakes Symposium on VLSI, GLSVLSI ’09, pages 315–320, New York, NY, USA, 2009. ACM.
[19]P. Mars and W.J Poppelbaum. Stochastic and deterministic averaging processors. pages 20–53.
Stevenage, Eng: P. peregrinus, 1981.
[20]Weikang Qian, M.D. Riedel, Hongchao Zhou, and J. Bruck. Transforming probabilities with com- binational logic. Computer-Aided Design of Integrated Circuits and Systems, IEEE Transactions on, 30(9):1279–1292, Sept 2011.
[21]N. Saraf and K. Bazargan. Sequential logic to transform probabilities. In Computer-Aided Design (ICCAD), 2013 IEEE/ACM International Conference on, pages 732–738, Nov 2013.
[22]Barry Shackleford, Motoo Tanaka, Richard J. Carter, and Greg Snider. Fpga implementation of neighborhood-of-four cellular automata random number generators. In Proceedings of the 2002 ACM/SIGDA Tenth International Symposium on Field-programmable Gate Arrays, FPGA ’02, pages 106–112, New York, NY, USA, 2002. ACM.
[23]S. Wolfram. Random Sequence Generation by Cellular Automata. Advances in Applied Mathe- matics, 7(2):267–307, 1986.
[24]Gu Xiao-chen and Zhang Min-xuan. Uniform random number generator using leap ahead lfsr architecture. In Computer and Communications Security, 2009. ICCCS ’09. International Con- ference on, pages 150–154, Dec 2009.