Abstract
Wireless Network-on-Chip (WNoC) is a promising paradigm to overcome the versatility and scalability issues of conventional on-chip networks for current processor chips. However, the chip environment suffers from delay spread which leads to intense Inter-Symbol Interference (ISI). This degrades the signal when transmitting and makes it difficult to achieve the desired Bit Error Rate (BER) in this constraint-driven scenario. Time reversal (TR) is a technique that uses the multipath richness of the channel to overcome the undesired effects of the delay spread. As the flip-chip channel is static and can be characterized beforehand, in this paper we propose to apply TR to the wireless in-package channel. We evaluate the effects of this technique in time and space from an electromagnetic point of view. Furthermore, we study the effectiveness of TR in modulated data communications in terms of BER as a function of transmission rate and power. Our results show not only the spatiotemporal focusing effect of TR in a chip that could lead to multiple spatial channels, but also that transmissions using TR outperform, BER-wise, non-TR transmissions it by an order of magnitude.
Introduction
As heterogeneous Systems-in-Package (SiP) integrate more CPUs, GPUs, and specialized accelerators, traditional wired interconnections face significant scalability and latency hurdles. Wireless Network-on-Chip (WNoC) has emerged as a high-bandwidth, low-latency alternative. However, the internal environment of a chip package acts as a complex electromagnetic enclosure, causing radio waves to reflect off metallic surfaces. This creates a high delay spread and severe Inter-Symbol Interference (ISI), where overlapping data bits become indistinguishable at high speeds. To achieve the 10+ Gb/s rates required for modern computing, a robust method to mitigate these reflections is essential.
Methodology and System Design
This research proposes the use of Time Reversal (TR) to turn the package's complexity into an advantage. In TR, the transmitter pre-codes the signal using a time-reversed version of the Channel Impulse Response (CIR). When this pre-distorted signal travels through the multipath environment, the reflections constructively interfere at the intended receiver, reconstructing a sharp, focused pulse.
The team characterized a 10x10 mm² flip-chip package at 60 GHz using the CST Microwave Studio full-wave electromagnetic solver. The model included realistic components such as the Aluminum Nitride heat spreader, silicon die, and copper solder bumps. Because the chip package is a static environment, the TR filter can be pre-calculated and hard-coded, avoiding the need for power-hungry real-time channel estimation hardware.
Spatiotemporal Focusing and Performance Results
A key finding of this study is the dual benefit of TR: temporal compression and spatial focusing. Temporally, TR mitigates ISI by focusing signal energy into a narrow window. Spatially, it concentrates energy at the target antenna, with the focusing peak amplitude reaching roughly three orders of magnitude higher than non-TR signals. This spatial isolation allows for potential spatial multiplexing, where multiple wireless links operate simultaneously on the same frequency without interference.
Performance evaluations using MATLAB demonstrated that TR-enabled systems can reach data rates of up to 60 Gb/s with ultra-low error rates. At a more standard 10 Gb/s, the system achieved a BER of $10^{-15}$ at 17dB SNR, matching the reliability standards of traditional wired interconnects. In contrast, non-TR transmissions typically see performance collapse beyond 5 Gb/s due to unmanaged ISI.
Conclusion
By leveraging Time Reversal, the chaotic reflections within a computer chip package can be harnessed to create high-speed, reliable wireless highways. While future work must address the physical area and power consumption of silicon-based TR filters, this research proves that TR is a viable path toward scalable, multi-gigabit wireless communication for next-generation computing architectures.
Summary updated