一种低输入要求的高精度随机计算乘法器

A high-precision stochastic computational multiplier with low input requirements

针对传统随机计算乘法输出精度受输入比特流之间相关性影响的问题,提出一种对相关性不敏感的高精度乘法器。首先,通过编码权重的转换和输入的组合,建立了乘法计算模型;其次,针对随机计算乘法电路的缩放输出问题,设计了非缩放优化电路并提高了计算精度;最后,分别使用多种分布比特流作为输入,验证了乘法器对比特流的适应情况。通过引入比特翻转并进行高斯滤波实验,验证了乘法器的有效性和容错性。仿真结果表明:与传统随机计算乘法器相比,该乘法器提高了54.8%的精度,节省了47.49%的面积;与二进制乘法器相比,节省了82.50%的面积。

To address the problem that the output accuracy of the traditional stochastic computation multiplication was affected by the correlation between the input bit streams,a high-precision multiplier insensitive to the correlation was proposed.Firstly,the multiplication computation model was established by converting the encoding weights and combining the inputs.Secondly,a non-scaling optimization circuit was designed to solve the scaling output problem of the stochastic computation multiplier circuit and improve the computational accuracy.Finally,several distributed bit streams were used as inputs to verify the adaptation of the multiplier to the bitstreams;the effectiveness and fault tolerance of the multiplier were verified by introducing bit flipping and performing Gaussian filtering experiments.The simulation results show that the multiplier outperforms the traditional stochastic computational multiplier in terms of accuracy increasement of 54.8%and area reduction of 47.49%.In comparison to the binary multiplier,the area is saved by 82.50%.