SAMPLING BASED RANDOM NUMBER GENERATOR SBRNG FOR STOCHASTIC
SAMPLING BASED RANDOM NUMBER GENERATOR (SBRNG) FOR STOCHASTIC COMPUTING M. Burak KARADENIZ & Mustafa ALTUN Emerging Circuits and Computation (ECC) Group Istanbul Technical University ICECS 2017
LITERATURE • LFSR • TRNG Problems • Speed • Feasibility • Uncontrollability • Lack in Randomness • Area Consumption
SAMPLING BASED RANDOM NUMBER GENERATOR (SBRNG)
BLOCK DIAGRAM
THEORY
PROPOSED CIRCUIT DIAGRAM Ø Signal Frequency and Amplitude are set by R 1, R 2, C 1, C 2
DESIGN FLOWCHART
ANALYSIS OF BIT STREAM COMPATIBILITY Ø HSPICE simulation is set and embedded into MATLAB code
UNIFORM DISTRIBUTION COMPATIBILITY Ø Sampling Errors from Single Sine Wave Sampling Based Random Number Generator (SBRNG) is not perfectly uniform distributed Ø Uniform Signals (such as Triangular Wave) can be used but yields poor randomness
UNIFORM DISTRIBUTION COMPATIBILITY
BİNOMİAL DİSTRİBUTİON COMPATIBILITY Ø SBRNG generated bit streams are compared to MATLAB’ s. Ø Probability Density Function (PDF) of SBRNG is binomial
PERFORMANCE of SBRNG Effective number of bits Transistor count with normalized% 50 probability output Transistor count with desired probability output LFSR 8 LFSR 13 MATLAB Randn (1, 160000) GLFSR [3] STT-MTJ [4] SRAM Based [6] All- Digital PVT Variation Tolerant [5] 256 8192 160000 256 256 1048576 8100 60000 114 174 NA 108 9719 1536 734693 61 81 101 290 460 NA Not reported 101 121 141 Non. Available (NA) CLK 0. 177 -0. 2 NA 0. 1 -2. 9 Clock Frequency (CLK) SBRNG 2 (this work) SBRNG 3 (this work) SBRNG 4 (this work) CLKx 100
CONCLUSION PROPOSED SBRNG’S HAVE A REAL POTENTIAL TO REPLACE LFSR BASED TECHNIQUES. Ø Proposed design gives 640 x higher throughput over conventional methods with enhanced simplicity of SBRNG network. Ø Speed of SBRNG is 100 x better compared to traditional LFSR (RNG speed is critical for security issues) Ø RNG stream probability can be controlled in SBRNG (very handful in Stochastic Applications)
FUTURE WORK ØVery Fast Highly Random Probability Controllable Sampling Based Random Number Generator will be Tape-out based on theoratical assumptions and simulations on this work.
THANK YOU FOR LISTENING. • This work is supported by the TUBITAK 1001 project # 116 E 250.
- Slides: 15