A novel, flexible circuit used to implement selected mathematical operations for AI algorithms optimized for hardware applications

The paper presents a novel circuit that enables an implementation of various mathematical operations with automatic conversion of the resulting multi-bit digital signals into voltage. The circuit consists of a number of parallel branches consisting of a combination of switches and resistors. Depending on the value of the resistive elements used, it enables the implementation of such operations as adding, subtracting and multiplying numbers stored as digital signals. Additionally, it can be used to obtain an approximation of selected nonlinear functions that include logarithmic, exponential, tangent functions. For binary weighted resistances (1, 2, 4, 8, …) in subsequent branches, the circuit works as a conventional voltage digital-to-analog converter. The functions that can be implemented with this circuit enable its application in various AI algorithms implemented as specialized integrated circuits. In this work, we focus in particular on its application to the implementation of basic operators of fuzzy logic systems. The circuit was implemented in CMOS 0.13μm technology.