Low power current-mode binary-tree asynchronous Min/Max circuit

Hierarchical Temporal Memory Based on Spin-Neurons and Resistive Memory for Energy-Efficient Brain-Inspired Computing

“…A winner-take-all (WTA) circuit receives the current inputs and determines the 'winner'. Several versions of WTA circuits have been proposed in literatures, which can be classified into two broad categories: current-conveyor WTA (CC-WTA) [26], and binary tree WTA (BT-WTA) [25,26], the latter being more suitable for large number of inputs [25,26]. BT-WTA employs a binary tree of 2-input comparison stages which involve copying and propagating the larger of the two current inputs to the output [25].…”

“…In order to maintain a sufficiently high resolution, larger transistor dimensions (both length as well as width) and hence, larger cell area is needed. This is evident from the recent designs [26] that although used relatively scaled CMOS technology, but inevitably employed significantly large channel lengths for such circuits. The increased parasitic capacitance coming from large channel lengths leads to lower operating frequency ( fig.…”

“…Higher frequency and resolution can be achieved at the cost of increased input currents, which lead to larger power consumption [24]. Special techniques to enhance the precision of current mirrors have been proposed in literature [26], but they introduce significant overhead in terms of power consumption and area complexity. Voltage-mode processing can also be employed in RCN, however, it incurs additional overhead due to current to voltage conversion and subsequent amplifications.…”

High-Performance Mixed-Mode Universal Min-Max Circuits for Nanotechnology

“…Minimum and Maximum functions, often indicated as Winner-Take-All (WTA) and Loser-Take-All (LTA), are used as the essential blocks in many applications such as image processing [45], min/max filtering [42], fuzzy systems [26,35], neural networks [14], half-wave and full-wave rectifiers and analog switches [37]. Fuzzy logic functions usually include three steps which are fuzzification, fuzzy rule evaluation, and defuzzification [35].…”

“…For instance, [26] presents a multi-input WTA/LTA circuit for the fuzzy applications. Moreover, realization of neural networks in analog circuits is preferable due to higher performance, smaller processing time and chip area [14].…”