Modeling and Design of Power Distribution Network for a Heterogeneous Integrated Active Interposer with Neuromorphic Computing Circuits

Ultra-high density 2.5/3D heterogeneous integration has been considered an essential solution for rebooting computation applications like high performance computing, machine learning, and brain-mimicking neuromorphic computing, in both cloud and edge modes. Interposers with active auxiliary circuitry such as tunable power distribution network (PDN), phase locked loops, active signaling equalizers/buffers, and even neuromorphic units, are emerging as much more attractive and flexible platforms than passive interposers for these applications. In this paper, an active interposer conception acting as a platform for heterogeneous integration of logic, memory and neuromorphic computing circuits, is proposed for rebooting computation purpose, featuring functioning units such as active PDN, data switching and signal conditioning circuits, and neuromorphic units based on so-called neural TSVs. New features such as pulse operation of neuron circuit and high speed data switching may induce new power integrity issues. Fortunately, the introduction of neural TSVs with large capacitance may also acting as the in-situ decoupling capacitors whose value can be modulated by the switching on/off of the MOSFETs associated with the neural TSVs in the same neuron cell, and thus a compact and tunable PDN can be constructed, whose impedance and anti-resonance characteristics may be reconfigured flexibly for an optimal power distribution and minimal simultaneous switching noise. Principles are disclosed and compact circuit modeling is set up for the PDN. Circuit and full-wave simulation results are demonstrated, confirming the conception and its effectiveness.