This promises to revolutionize how chiplets communicate—offering faster, more efficient, and scalable computing solutions.
As computing technology evolves, researchers are moving away from large, single-chip processors to systems composed of smaller, specialized components known as “chiplets.” These chiplets collaborate to enhance processing power and efficiency. This shift has become essential as the industry reaches the physical limits of fitting more transistors on a single chip, leading to challenges like overheating and power inefficiency. By using multiple chiplets, computing power can be increased without hitting these constraints.
A major challenge in this transition is efficient communication between chiplets. Historically, this was managed by Network-on-Chip (NoC) systems, which act like data highways within a chip. However, as chiplet systems become more complex, this wired approach creates delays, higher energy consumption, and scalability issues. In larger systems, this evolves into Network-in-Package (NiP), but it faces the same limitations as data must travel across increasingly vast grids.
To address this, a team of researchers is investigating wireless communication at the chip level. Rather than using wired connections, chiplets could communicate wirelessly via tiny antennas using terahertz (THz) frequencies. These frequencies, positioned between infrared and microwave signals, promise high-speed data transfer. However, the sensitivity of THz signals to noise complicates this process, leading to decoding difficulties. The research team applied Floquet engineering, a quantum physics technique, to solve this problem. This approach controls electron behavior when exposed to high-frequency signals, making the system more responsive to certain frequencies and better at detecting THz signals amidst noise. The study, conducted using a two-dimensional semiconductor quantum well (2DSQW).
The researchers also developed a dual-signaling architecture, where two receivers adjust the system’s reference voltage in real time based on detected noise levels. This setup significantly reduces error rates compared to single-receiver systems, ensuring more accurate signal decoding. The team mentioned that their approach marks a major advancement towards reliable, high-speed wireless communication between chiplets, paving the way for more efficient and scalable computing systems.
