Towards spatial multiplexing in wireless networks within computing packages

Fátima Rodríguez-Galán, Elana Pereira de Santana, Peter Haring Bolívar, Sergi Abadal, Eduard Alarcón

TL;DR

Researchers have demonstrated that it is possible to create multiple independent wireless "lanes" inside a single computing package using tiny antenna arrays. By precisely controlling the phase of signals at 60 GHz and 110 GHz, they achieved a massive 40 dB signal-to-interference ratio, effectively paving the way for high-bandwidth, low-latency wireless communication between chiplets.

What Problem Does This Paper Address?

As we move into the era of "manycore" processors and heterogeneous "chiplet" designs, the physical wires connecting these components (Networks-on-Chip or NoCs) are becoming a major performance bottleneck. Wireless Networks-on-Chip (WNoCs) have been proposed as a solution to provide fast, long-range communication across a chip.

However, WNoCs face a "bandwidth crunch." Because all antennas typically share the same frequency space within the tiny, enclosed environment of a chip package, they usually can't talk at the same time without interfering. This paper tackles the challenge of spatial multiplexing: how to create multiple, concurrent, and non-interfering wireless channels within the cramped, reflective confines of a flip-chip package.

Methods

The authors conducted a feasibility study using high-fidelity "full-wave" electromagnetic simulations (via CST Microwave Studio).

  • The Environment: They modeled a standard flip-chip package, including the silicon die, solder bumps, and a heat spreader.
  • The Antennas: Instead of traditional horizontal antennas, they used vertical monopoles implemented as Through-Silicon Vias (TSVs). These are compact and integrate well with existing chip manufacturing.
  • The Array: They designed a 4x4 array of these monopoles. By adjusting the excitation phase (the timing) of each antenna element—a technique known as beamforming—they attempted to concentrate the electromagnetic field in specific spots on the chip.
  • The Strategy: Unlike previous work that tried to use external "reflectarrays" to bounce signals, this approach relies on reconfigurable intelligence at the transmitter and receiver ends to create isolated spatial channels.

Datasets

As this is a computational electromagnetics study, no external datasets were used. The results are derived from physical simulations based on the material properties of silicon, copper, and aluminum nitride (the heat spreader) at millimeter-wave frequencies.

Key Insights and Ideas

  • Silicon as an Asset: While silicon is often seen as a "lossy" medium for radio waves, the researchers found that these losses actually help. The high permittivity of silicon allows for much smaller antenna arrays (the 4x4 array fits in just 1.18 mm²), and the material's natural attenuation helps dampen interference between adjacent spatial channels.
  • The "Enclosure" Effect: In a typical wireless setting, reflections are a nightmare. Inside a chip package, the researchers found that the proximity of the package "walls" and the ground plane (the solder bump layer) actually helps in shaping the beams toward the center of the chip.
  • Phase is Everything: The study shows that simply by changing the phase distribution across the 16 elements, they could steer the energy to different corners of the chip at will, creating "virtual wires" in the air.

Key Results

  • High Isolation: At 60 GHz, the system created two parallel channels with a Signal-to-Interference Ratio (SIR) of 40 dB. In wireless terms, this is exceptionally clean, proving the channels are almost entirely independent despite being millimeters apart.
  • Frequency Scaling: The approach successfully scaled to 110 GHz. At this higher frequency, the antennas become even smaller, and the S11 parameters (a measure of antenna efficiency) remained favorable.
  • Triple Channels: The team demonstrated the feasibility of supporting three concurrent spatial channels by placing arrays in different configurations, showing that the "spatial real estate" inside a chip is larger than previously thought.
  • Compact Footprint: The 16-element array occupies only about 1.18% of a standard 10x10 mm² chip, making it a practical addition to modern processor designs.

Limitations and Future Work

  • Simulation vs. Reality: The primary limitation is that this is a simulation-based study. Real-world factors like manufacturing tolerances, thermal fluctuations, and the presence of complex metal routing layers in a real chip could impact performance.
  • Hardware Complexity: Implementing high-speed, on-chip phase shifters for a 16-element array is a significant circuit design challenge that wasn't the focus of this paper.
  • Next Steps: The authors suggest expanding the approach to frequencies beyond 100 GHz (sub-terahertz) and exploring even more concurrent channels to further boost the aggregate bandwidth of future computing systems.

Summary updated

Abstract

Wireless Networks-on-Chip (WNoCs) are regarded as a disruptive alternative to conventional interconnection networks at the chip scale, yet limited by the relatively low aggregate bandwidth of such wireless networks. Hence, any method to increase the amount of concurrent channels in this scenario is of high value. In this direction, and since WNoC implies close integration of multiple antennas on a chip anyway, in this paper we present a feasibility study of compact monopole antenna arrays in a flip-chip environment at millimeter-wave and sub-terahertz frequencies. By means of a full-wave solver, we evaluate the feasibility to create, at will, concentrations of field in different spots of the chip. This way, we set the steps towards spatial multiplexing that enables concurrent multicast communications and also increases the aggregate bandwidth of the wireless network. Our results at 60 GHz show two clearly separable parallel channels that radiate simultaneously from two opposite corners of the chip, achieving a Signal-to-Interference Ratio (SIR) of around 40 dB, which proves that the channels are independent of each other even in such an enclosed environment. Further, we see potential to expand our approach to three or more concurrent channels, and to frequencies beyond 100 GHz.
BibTeX
@inproceedings{rodriguezgalan2022towards,
  author = {Fátima Rodríguez-Galán and Elana Pereira de Santana and Peter Haring Bolívar and Sergi Abadal and Eduard Alarcón},
  title = {{Towards spatial multiplexing in wireless networks within computing packages}},
  booktitle = {NANOCOM},
  year = {2022},
  url = {https://openreview.net/forum?id=4ubbWAFCRc},
}

Loading related content...

Figure