摘要
提出了可构成径向基函数(RBF)神经网络的CMOS模拟单元电路,包括绝对值电路、电流求均方根电路和类高斯函数电路.基于这些电路设计了一个二输入/一输出,含有两个隐层神经元的径向基函数神经网络,并通过异或问题进行了验证.所有单元均采用HJTC0.18μmCMOS数模混合工艺制造,芯片面积为200μm×150μm,功耗约为100μW.芯片测试结果表明:提出的单元电路结构简单,功耗低,便于扩展和调节,因而为硬件实现径向基函数神经网络的片上学习提供了可能.
Several kernel CMOS circuits for radial basis function （RBF） networks are proposed, including current type absolute circuit, mean square root circuit, and an adjustable Gaussian circuit. A two-input/one-output RBF network with two hidden nodes composed of these circuits is implemented and verified for the XOR problem. All the circuits were fabricated in HJTC 0.18μm CMOS technology. The chip area is 200μm ×150μm and the power consumption is about 100μW. Experimental results show that all the proposed circuits provide simple structures, low power consumption, and high operation capacity. This demonstrates the possibility of a hardware implementation for on-chip learning in an RBF network.
