# Why Chips Are Using On-Chip Antennas for Wireless Interconnects

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 28 Jun 2026, 7:59.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0078.mp3

In this episode of The Hardware Podcast, Lucas and Luna explore the emerging trend of integrating antennas directly onto silicon chips to replace traditional wired interconnects for short-range data communication. They discuss a 2025 paper from Intel Labs demonstrating a 60GHz on-chip antenna array achieving 40 Gbps data rates over a few millimeters, the engineering challenges of antenna efficiency in silicon, and how this technology could reshape chip-to-chip communication in data centers and multi-chip modules. The conversation also covers comparisons with existing solutions like TSV and optical interconnects, and the potential for wireless interconnects to reduce latency and power consumption in dense compute clusters. Tune in for a detailed look at a cutting-edge hardware innovation that may soon become standard in high-performance computing. #OnChipAntennas #WirelessInterconnects #ChipDesign #IntelLabs #60GHz #SiliconPhotonics #DataCenter #Chiplet #HighPerformanceComputing #InterconnectTechnology #Technology #HardwareEngineering #FexingoBusiness #BusinessPodcast #Semiconductors #WirelessCommunication #MMWave #AntennaDesign Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So there's a paper from Intel Labs published late last year that I keep coming back to. It's about putting tiny antennas directly onto the silicon die — not for WiFi or Bluetooth, but for communicating between chips just millimeters apart. They demonstrated a 60 gigahertz antenna array that achieved 40 gigabits per second over a distance of a few millimeters.
Luna: Forty gigabits per second over a few millimeters — that's essentially on-package distances. Why would you want wireless for something that close? Isn't a wire simpler?
Lucas: It's a fair question. Wires are simple until you have hundreds of them. In a multi-chip module or a data center server, the number of interconnects grows faster than the bandwidth each one can carry. You end up with signal integrity issues, crosstalk, power loss, and routing congestion. Wireless interconnects can cut through that — literally.
Luna: Right — no physical trace means no impedance matching headaches, no parasitic capacitance. But integrating an antenna on silicon is not trivial. Silicon is lossy at high frequencies.
Lucas: Exactly. At 60 gigahertz, the skin depth in silicon is only a few micrometers. The Intel team got around that by using a thick oxide layer under the antenna to isolate it from the lossy substrate. They also used a technique called 'lens integration' — a small dielectric lens placed above the antenna to focus the beam.
Luna: So the antenna itself is on the chip, but the lens is external? That still requires assembly.
Lucas: In that prototype, yes. But the lens is a passive component — cheap and easy to attach. The key innovation is that the active circuitry — the oscillator, the modulator, the antenna feed — is all on the same CMOS die. That eliminates the need for a separate radio chip and the associated interconnect.
Luna: And 40 Gbps per link — how does that compare to wired alternatives like through-silicon vias or micro-bumps?
Lucas: TSVs and micro-bumps can achieve similar or higher bandwidth, but they require physical connection — which means you need precise alignment, and you're limited by the pitch of the bumps. Wireless interconnects don't need physical contact, so alignment tolerance is much looser. That translates to higher yield in manufacturing.
Luna: So this could be a game-changer for chiplet architectures. Right now, chiplets talk to each other through a silicon interposer with thousands of wires. But that interposer is expensive to fabricate.
Lucas: Exactly. Wireless interconnects could eliminate the interposer altogether. You just place the chiplets close together on a package substrate, and they communicate via these on-chip antennas. Intel's paper showed a power efficiency of about 2 picojoules per bit — competitive with wired links at similar data rates.
Luna: Two picojoules per bit — that's actually pretty good. For comparison, a typical SerDes link — serializer-deserializer — can be around 5 to 10 picojoules per bit.
Lucas: Right. And the latency is lower because you don't have the serialization delay. The antenna array can operate in parallel, so you get a wide data bus without the routing overhead.
Luna: But there's a catch, isn't there? At 60 gigahertz, the wavelength is about 5 millimeters in free space. For effective transmission, the antennas need to be a certain fraction of that wavelength, which means they occupy significant die area.
Lucas: That's the trade-off. In Intel's design, each antenna element was about 1.2 by 1.5 millimeters — that's not tiny. For a 16-element array, you're looking at a significant chunk of silicon. But if you're replacing a wide bus of hundreds of wires, the area might actually be comparable.
Luna: And there's also the question of interference. If every chiplet is broadcasting in the same frequency band, how do you prevent collisions?
Lucas: The paper addressed that by using directional beams and time-division multiplexing. Each antenna array is steered electronically — phased-array style — so you can create point to point links that don't interfere with each other. In a controlled environment like a package, the paths are predictable.
Luna: So this is not WiFi for your laptop — this is a highly engineered, short-range, directed link. It's almost like a wired connection but without the wire.
Lucas: Exactly. And the potential applications go beyond chiplet packages. Think about a server rack where blades are only a few centimeters apart. You could have wireless backplane communication. Or in a data center, top of rack switches talking to servers without cables.
Luna: That would be huge for airflow and cooling — no cables to block airflow.
Lucas: Right. And for reconfigurable systems — you could dynamically reassign links between compute nodes on the fly. That's hard to do with physical cables.
Luna: So where does this technology stand today? Is it just a lab curiosity, or are there commercial products on the horizon?
Lucas: It's still mostly research. Intel's paper is from 2025 — it's a proof of concept. But there are startups working on wireless interconnects for data centers. One company called 'SiBeam' has been doing 60 GHz wireless for consumer devices, but the chip to chip space is still nascent. The big foundries — TSMC, Samsung — are starting to offer process design kits that include antenna models, so it's becoming feasible.
Luna: That's a strong signal — if the foundries are supporting it, they expect demand. It reminds me of the early days of silicon photonics, when people said it would never be practical.
Lucas: Exactly. And look where silicon photonics is now — it's being deployed in data centers for long-haul links. On-chip antennas could follow a similar trajectory, but for short-reach links.
Luna: I think this is one of those technologies that will quietly become ubiquitous. We'll start seeing it in high-end server chips first, then maybe trickle down to consumer devices.
Lucas: I agree. And that's why we cover it here — because it's the kind of hardware engineering that doesn't get headlines but fundamentally changes how systems are built.
Luna: Speaking of what we cover here — we do this show without any ads or sponsors. That means no annoying mid-roll interruptions. If today's conversation gave you something useful, and you want to support that ad-free approach, it really helps. There's a link at buy me a coffee dot com slash fexingo.
Lucas: Yeah, it's a small way to keep the lights on and keep the content focused on the engineering, not on selling stuff. We appreciate anyone who chips in.
Luna: So back to antennas — what's the one thing you think listeners should take away from this episode?
Lucas: That the idea of a wireless interconnect is not science fiction — it's already been demonstrated at competitive speeds and power. The main barriers are cost and integration complexity, but those are being chipped away. Within five years, I'd expect to see commercial products using on-chip antennas for chip to chip communication.
Luna: And for listeners who want to dig deeper, what should they search for?
Lucas: Search for '60 GHz on-chip antenna array CMOS' — you'll find Intel's paper and also some work from imec and the University of Tokyo. The field is still small, so it's easy to catch up on.
Luna: I'll add that to my reading list. Great discussion today.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.