The Research And Implementation Of Random Numbers Accelerator Based FPGA For Monte Carlo

High quality random numbers are widely applied in the domains ofscientific applications. One of the extremely important instances is theMonte-Carlo simulation, which has been widely used in particle transport,computational physics, quantum thermodynamic calculations andaerodynamic calculations.Compared with other methods, FPGA based random numbergeneration exhibits much better performance on speedup, powerdissipation and cost. In this paper, we mainly use the techniques of FPGAto accelerate the parallelism generation of random numbers.The majorcontributions of this thesis are:1. We propose a hardware architecture called MTPRNG (MersenneTwister Parallel Random Number Generator). Compared with otherreported architecture, MTPRNG has much better performance on hardwareconsumption.2. In order to achieve the reconfigurable random number generationaccelerator, we have a deep research in Fast-jump-ahead algorithm.According to the characteristics of Mersenne Twister algorithm, weimplemented the Fast-jump-ahead algorithm in software.3. We propose a reconfigurable Software/Hardware framework forMT19937, which is capable of generating long-period random numbers inparallel. The hardware part of this framework is consisted of a parallelizedMTPRNG array according to the requirement of the application. Thesoftware part of this framework is composed of User Interface andFast-jump-ahead algorithm. With the support from software, this framework can generate random numbers at any degree of parallelism.Therefore, it can achieve orders of magnitude improvement in performancecompared to the dedicated software version.4. We apply our framework to a Monte-Carlo simulation forestimating the value of π. Experimental results prove the correctness ofour framework.