论文部分内容阅读
为了解决传统的多比特数字方法在实现人工神经网络时所存在的硬件规模过于庞大的问题,提出了一种基于总和增量(ΣΔ)调制的比特流人工神经网络硬件实现方法.在FPGA中实现了比特流加法器、乘法器、阈值函数单元和全数字ΣΔ调制器,并采用这些比特流运算单元构建了比特流人工神经元.用实现逻辑函数功能和线性分类器的方法验证了比特流人工神经元的功能.基于比特流人工神经元的带预处理结构的比特流感知器被证明具有非线性分类的能力.比特流人工神经元实现所使用的FPGA资源表明比特流人工神经网络硬件实现技术可以显著地减少人工神经网络对硬件资源的需求.
In order to solve the problem that the hardware scale of traditional multi-bit digital method is too large when implementing artificial neural network, a hardware implementation method of bit-stream artificial neural network based on total delta sigma modulation is proposed. Bit stream adder, multiplier, thresholded function unit and full digital ΣΔ modulator are used to construct the bit-stream artificial neural network.The bit-stream artificial neural network is constructed by using the logic function and linear classifier Neuronal Functions Bitstream-based biosensors based on bit-stream artificial neurons with a preconditioning structure proved to have the ability to be non-linearly categorized. The FPGA resources used by the bit-stream artificial neurons indicate that the bit-stream artificial neural network hardware implementation technique Can significantly reduce the artificial neural network hardware resources demand.