💻 computer science

Spying Across Chiplets: Side-Channel Attacks in 2.5/3D Integrated Systems

This paper demonstrates that side-channel attacks can be executed across chiplets in 2.5D/3D integrated systems by repurposing communication-oriented interfaces, such as antennas or RFID elements, to capture signals correlated with the activity of neighboring victim chiplets.

Original authors: Giorgio Di Natale, Christelle Rabache, Pierre-Louis Hellier, Florence Podevin, Sylvain Bourdel, Romain Siragusa, Paolo Maistri

Published 2026-05-12

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Giorgio Di Natale, Christelle Rabache, Pierre-Louis Hellier, Florence Podevin, Sylvain Bourdel, Romain Siragusa, Paolo Maistri

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a high-tech city built inside a single, tiny box. In this city, instead of having one giant building where everything happens, the work is split up among several smaller, specialized houses called chiplets. One house does the math, another handles memory, and a third is designed specifically to talk to the outside world using radio waves (like a walkie-talkie or an RFID tag).

The paper you shared is about a sneaky way a thief could steal secrets from the "math house" by listening in through the "radio house," even though they are supposed to be separate.

Here is the breakdown of their discovery:

1. The Setup: Neighbors in a Tiny Apartment

Think of the chip package as a very crowded apartment building.

The Victim: A chiplet doing secret math (like cracking a password). Every time it does a calculation, it makes tiny electrical "noises" or vibrations, just like a person talking in a room.
The Spy: A neighboring chiplet designed to send and receive radio signals. It has an antenna, which is basically a metal stick designed to catch waves from the outside world.

2. The Big Idea: The "Dual-Use" Antenna

Usually, we think of an antenna as a tool for talking to the outside. But the researchers realized that an antenna is also a great tool for listening to your neighbors.

Because the "math house" and the "radio house" are packed so tightly together inside the same package, the tiny electrical vibrations from the secret math can jump across the gap and vibrate the antenna of the radio chiplet.

The paper claims that a hacker doesn't need to break into the package or stick a probe on the victim chip. Instead, they can just take control of the "radio house" (the communication chiplet) and use its antenna to eavesdrop on the "math house."

3. How the Attack Works (The Experiment)

The researchers tested this idea in three steps:

Step 1: The Direct Listen. First, they listened directly to the "math house" using a standard tool. They successfully figured out the secret key just by hearing the electrical noise. This proved the noise was loud enough to be useful.
Step 2: The Simulation. Next, they used a computer to simulate the "radio house" sitting next to the "math house." They modeled how the electrical noise would travel through the air (or the material between them) and hit the antenna. They found that the noise did jump across, but it got distorted, like a voice sounding different when heard through a wall.
Step 3: The Recovery. Finally, they tried to decode the secret using the distorted signal caught by the "radio house."
- The Twist: The signal wasn't a perfect copy. It was changed by the physics of how it traveled (like how a sound changes when it bounces off a wall).
- The Fix: For one type of connection (capacitive), the signal looked like the speed of the noise rather than the noise itself. The researchers had to do a little math "reverse engineering" (integration) to turn it back into a readable message.
- The Result: Even with the distortion, they could still recover the secret key.

4. What This Means

The paper concludes that physical isolation isn't enough. Just because two chips are logically separated (they don't share data files) doesn't mean they are physically isolated.

The Metaphor: Imagine you are in a soundproof room trying to keep a secret. You think you are safe. But your neighbor, who is supposed to be listening to the radio, has a microphone that is so sensitive it picks up your heartbeat through the floorboards. If the neighbor is a spy, they can hear your secret without ever entering your room.

Summary of Findings

The Threat: A malicious or compromised communication chiplet can act as a spy for its neighbors.
The Method: It uses the chiplet's own antenna or contactless parts to pick up electromagnetic "leaks" from a nearby chiplet.
The Reality: The signal gets messy and changes shape as it travels, so the attacker has to be smart about how they clean up the noise to read the secret.
The Limitation: The researchers admit their computer simulation was a simplified model. Real-world chips might have more noise or different materials that make the signal weaker, but the basic principle holds: the signal can leak.

In short, the paper warns that in the future of tiny, stacked computer chips, the parts designed to talk to the outside world might accidentally (or intentionally) become the best listening devices for spying on your secrets.

Technical Summary: Spying Across Chiplets: Side-Channel Attacks in 2.5/3D Integrated Systems

Problem Statement
The semiconductor industry is shifting from monolithic scaling to advanced packaging and chiplet-based architectures (2.5D/3D integration) to achieve modularity, yield improvements, and performance gains. However, this transition introduces new physical attack surfaces. While prior research has established that logical isolation does not guarantee side-channel isolation in shared FPGA platforms or multi-chip boards, the specific security implications of heterogeneous chiplet integration remain under-explored.

The core problem addressed is whether a communication-oriented chiplet—legitimately integrated into a package to interact with the external environment via antennas, RFID-like elements, or contactless coupling structures—can be repurposed as an internal observation platform. The authors investigate if such a chiplet can capture electromagnetic or contactless leakage from a neighboring "victim" chiplet executing security-sensitive workloads (e.g., cryptographic operations), thereby enabling side-channel attacks from within the package without external probes.

Methodology
The paper proposes a threat model where an adversary controls or misuses a legitimate, RF-capable communication chiplet to spy on a co-packaged victim. The attack relies on the physical proximity of the chiplets within the same package or stack, exploiting coupling mechanisms such as electromagnetic emissions, near-field coupling, substrate coupling, or package-level propagation.

The experimental validation follows a three-step flow to assess feasibility:

Baseline Validation: A Differential Power Analysis (DPA) attack is performed on a 130 nm cryptographic circuit (AES S-box operation) using direct supply current measurements ( $I(V_{DD})$ ) to confirm the presence of exploitable, secret-dependent leakage.
Electromagnetic Simulation: A 3D electrical simulation using Finite Element Method (FEM) solvers models the coupling between the victim chiplet and an observing chiplet. Two coupling scenarios are simulated:
- Capacitive Coupling: Modeled via a metal plate probe.
- Inductive + Capacitive Coupling: Modeled via a metal line probe aligned with the power supply rail.
  The simulation uses a custom material stack to emulate the inter-chiplet medium (air gaps or dielectric) and captures the leakage signal induced on the observer.
Signal Analysis: The coupled signals retrieved on the observing chiplet are analyzed using the same DPA methodology. For capacitive coupling, the raw signal (which resembles the derivative of the original current) is numerically integrated to reconstruct a waveform compatible with the standard Hamming weight leakage model. For inductive+capacitive coupling, the raw signal is analyzed directly.

Key Contributions

Threat Model Definition: The paper formalizes a realistic adversary model where a communication chiplet acts as a dual-use structure: functioning as intended for external communication while simultaneously serving as an internal side-channel sensor.
Feasibility Demonstration: Through experimental simulation, the authors demonstrate that secret-dependent information (specifically, an 8-bit AES key) can be recovered from a neighboring chiplet via cross-chiplet coupling.
Signal Characterization: The study reveals that the leakage observed across chiplets is not a simple replica of the victim's signal. The coupling mechanism significantly transforms the waveform (e.g., acting as a high-pass filter in capacitive coupling), necessitating specific signal reconstruction or adaptation strategies before successful key recovery.

Results

Baseline: The DPA attack on the direct supply current successfully recovered the secret key using 256 traces.
Capacitive Coupling: The signal captured by the observing chiplet was the temporal derivative of the original current. After numerical integration to reconstruct the original waveform, the DPA attack successfully recovered the secret key, showing a dominant peak for the correct hypothesis.
Inductive + Capacitive Coupling: The raw coupled signal contained exploitable information, but the leakage was less distinct in the early parts of the trace due to complex propagation effects. The secret key could still be recovered, but the distinguisher peak was primarily visible during the final switching events of the sensitive operation.
Limitations: The authors note that the simulation used simplified material stacks and geometries that do not perfectly reflect actual Back-End-Of-Line (BEOL) technologies or realistic power delivery networks (e.g., decoupling capacitors). Consequently, while the results prove signal leakage is possible, the power and integrity of the leaked signal in a real-world scenario require further caution.

Significance and Claims
The paper claims to fill a critical gap in the literature by experimentally investigating the repurposing of communication interfaces for side-channel spying within a single package. It argues that the boundary between communication structures and sensing structures in heterogeneous systems is becoming blurred.

The primary significance lies in demonstrating that physical isolation at the logical level does not imply side-channel isolation at the system level in advanced packaging. The authors conclude that future secure designs for multi-chiplet architectures must account for the dual-use risk of communication interfaces and the physical coupling effects inherent in 2.5D/3D integration. They suggest that effective countermeasures will require addressing physical leakage paths created by package-level coupling and the relative placement of heterogeneous chiplets, rather than relying solely on logical isolation.