TNN7: A Custom Macro Suite for Implementing Highly Optimized Designs of Neuromorphic TNNs

Cortical Columns Computing Systems: Microarchitecture Model, Functional Building Blocks, and Design Tools

“…We also derived a set of characteristic equations from the gate-level designs that provide qualitative estimation of gate-level PPA complexity for arbitrary mini-column designs. functional building blocks [12] and design exploration tools. Some applications of interest are visual object recognition, anomaly detection [13], keyword spotting, and human activity recognition [56].…”

“…Design space exploration for UCR time-series clustering: Using TNN7 macros, PPA [12] for seven sample TNN prototype designs (mini-columns) across diverse synapse counts and application benchmarks from [13].…”

“…From the TNN microarchitecture model, we identified the key fundamental building blocks and implemented them as a suite of custom standard cell macros in 7 nm CMOS, called TNN7 [12]. Using these custom macros, we can further optimize the TNN designs and enhance their scalability.…”

“…Further, instantiating the TNN7 macros during logic synthesis reduces the netlist generation runtime considerably (by more than 3x typically). Functional building blocks of a mini-column (q neurons with p synapses each) and associated proposed TNN7 macros (highlighted in yellow) [12]. Two macros are developed for synaptic feedforward processing, three for synaptic STDP learning, one for WTA inhibition and three for generic utility functions such as spike encoding.…”

“…TNNs encode and process information in temporal form mimicking the brain's neocortical sensory signal processing. We developed a microarchitecture model for implementing highly efficient TNN designs [11,12] and demonstrated state-of-the-art clustering performance on a wide variety of time-series signals [13].…”