3D Wireless Channel Modeling for Multi-layer Network on Chip

TL;DR

This paper introduces novel 3D wireless channel models for multi-layer network on chip systems, enabling efficient and accurate characterization of complex wireless paths with minimal computational cost.

Contribution

It proposes boundary-less and boundary-constrained 3DWiNoC models with an efficient approximation algorithm for accurate channel gain estimation in multi-layer chips.

Findings

Difference rate between models is below 0.001% for 20 layers.

Channel gain decreases sharply with increased refract time.

Approximate algorithm achieves error rate below 0.1%."

Abstract

The resource constraints and accuracy requirements for Internet of Things (IoT) memory chips need three-dimensional (3D) monolithic integrated circuits, of which the increasing stack layers (currently more than 176) also cause excessive energy consumption and increasing wire length. In this paper, a novel 3D wireless network on chips (3DWiNoCs) model transmitting signal directly to the destination in arbitrary layer is proposed and characterized. However, due to the the reflection and refraction characteristics in each layer, the complex and diverse wireless paths in 3DWiNoC add great difficulty to the channel characterization. To facilitate the modeling in massive layer NoC situation, both boundary-less model boundary-constrained 3DWiNoC model are proposed, of which the channel gain can be obtained by a computational efficient approximate algorithm. These 3DWiNoC models with approximation algorithm can well characterize the 3DWiNoC channel in aspect of complete reflection and refraction characteristics, and avoid massive wired connections, high power consumption of cross-layer communication and high-complexity of 3DWiNoC channel characterization. Numerical results show that: 1) The difference rate between the two models is lower than 0.001% (signal transmit through 20 layers); 2) the channel gain decreases sharply if refract time increases; and 3) the approximate algorithm can achieve an acceptable accuracy (error rate lower than 0.1%).