Abstract
Wireless Network-on-Chip (WNoC) appears as a promising alternative to conventional interconnect fabrics for chip-scale communications. WNoC takes advantage of an overlaid network composed by a set of millimeter-wave antennas to reduce latency and increase throughput in the communication between cores. Similarly, wireless inter-chip communication has been also proposed to improve the information transfer between processors, memory, and accelerators in multi-chip settings. However, the wireless channel remains largely unknown in both scenarios, especially in the presence of realistic chip packages. This work addresses the issue by accurately modeling flip-chip packages and investigating the propagation both its interior and its surroundings. Through parametric studies, package configurations that minimize path loss are obtained and the trade-offs observed when applying such optimizations are discussed. Single-chip and multi-chip architectures are compared in terms of the path loss exponent, confirming that the amount of bulk silicon found in the pathway between transmitter and receiver is the main determinant of losses.
Introduction
As the industry moves toward heterogeneous multi-chip architectures, such as 2.5D and 3D integration, traditional wired interconnects are struggling to meet the escalating latency and bandwidth demands of communication between CPUs, GPUs, and memory. Wireless Network-on-Chip (WNoC) has emerged as a promising alternative, offering low-latency broadcast capabilities without the complexities of physical routing.
However, a significant gap exists in current WNoC research: the wireless channel is often treated as an abstraction, ignoring the physical reality of the chip package. In practice, a computing package is a dense sandwich of materials—including lossy silicon, metallic heat sinks, and ceramic carriers—that profoundly affects millimeter-wave propagation. This paper provides a rigorous electromagnetic characterization of these channels to prevent "overly optimistic" transceiver designs.
Modeling and Methodology
The researchers employed full-wave Electromagnetic (EM) simulations (via CST Microwave Studio) to model a variety of flip-chip package configurations. The modeling included the entire physical stack: metallic heat sinks, Aluminum Nitride (AlN) heat spreaders, silicon dies, interconnection layers ($SiO_2$), solder bumps, and the interposer or ceramic carrier.
A key methodological innovation was the exploration of Through-Silicon Vias (TSVs) as vertical monopole antennas. Unlike standard planar antennas, these radiate laterally, making them ideal for intra-package communication. The study also involved joint antenna-package optimization, varying the thickness of the silicon die and the heat spreader to identify optimal propagation conditions.
Key Insights and Results
The Impact of Package Materials
- Silicon as an Obstacle: Bulk silicon ($10 \Omega \cdot cm$) is highly lossy. The study found that thinning the silicon die from 0.7 mm to 0.1 mm can improve signal coupling by over 40 dB. The volume of bulk silicon in the direct path is the primary determinant of signal loss.
- Heat Spreader as a Waveguide: While primarily a thermal component, a thick AlN heat spreader acts as a low-loss medium. Increasing its thickness can improve coupling by up to 33 dB compared to configurations without a spreader.
The Waveguiding Effect
Inside a package, metallic walls act like a "micro-reverberation chamber." These reflections keep energy contained, creating a waveguiding effect that reduces path loss. In an optimized single-chip package, the authors achieved a path loss exponent ($n$) as low as 0.75, which is significantly more efficient than the $n=2$ found in free space. Conversely, removing the metallic package walls increased worst-case link loss by up to 27 dB.
Comparative Architecture Analysis
The propagation path varies significantly across different integration strategies:
- Silicon Interposers (2.5D): Waves can travel through vacuum or air gaps between chips, resulting in a relatively low path loss exponent of $n=1.55$.
- Multi-Chip Modules (MCM): Waves are often forced through more lossy silicon, causing the exponent to rise sharply to $n=4.27$.
Conclusion and Future Outlook
This characterization demonstrates that with careful co-design of the package and antennas, wireless links can achieve path loss exponents significantly lower than those of free space. However, challenges remain. Optimal package dimensions are frequency-dependent, and using TSVs as antennas requires non-standard fabrication techniques. Future work must also address the thermal impacts of extreme silicon thinning and the potential for interference between multiple concurrent wireless links within the enclosed package environment.
Summary updated