---
type: Scholarly Article
title: "TraceFormer: A Transformer-Based Method for Weight Extraction from AIMC Tiles"
description: Shows that the weights stored in an analog in-memory computing tile can be recovered from the power consumption of its ADCs, using a Transformer and an input-controlled weight-isolation technique.
resource: https://doi.org/10.1109/DSD67783.2025.00092
tags:
  - analog in-memory computing
  - hardware security
  - power side channels
  - transformers
  - weight extraction
  - vulnerabilities
  - power side-channel analysis
  - adc power analysis
generated:
  by: process:roozbeh.nl/astro-build
  at: 2026-09-06T00:00:00Z
status: stable
sources:
  - id: crossref
    resource: https://api.crossref.org/works/10.1109/DSD67783.2025.00092
    title: Crossref metadata record
  - id: publisher
    resource: https://doi.org/10.1109/DSD67783.2025.00092
    title: Publisher record (IEEE)
  - id: pdf
    resource: https://roozbeh.nl/papers/TraceFormer.pdf
    title: Author manuscript (PDF)
doi: 10.1109/DSD67783.2025.00092
authors:
  - name: Roozbeh Siyadatzadeh
    orcid: 0000-0002-2258-4110
  - name: Fatemeh Mehrafrooz
    orcid: 0000-0001-5844-3793
  - name: Nele Mentens
    orcid: 0000-0001-8753-7895
  - name: Todor Stefanov
    orcid: 0000-0001-6006-9366
year: 2025
published: 2025-09-10
venue: 2025 28th Euromicro Conference on Digital System Design (DSD)
venue_type: conference
publisher: IEEE
pages: 635-642
web_page: https://roozbeh.nl/publications/traceformer-weight-extraction-aimc/
pdf: https://roozbeh.nl/papers/TraceFormer.pdf
pdf_version: author
bibtex: https://roozbeh.nl/publications/traceformer-weight-extraction-aimc.bib
funding:
  - Horizon Europe project NeuroSoC (European Union, grant 101070634)
citation_count: 0
citation_count_source: Semantic Scholar Graph API, 2026-09-02
---

# TraceFormer: A Transformer-Based Method for Weight Extraction from AIMC Tiles

Roozbeh Siyadatzadeh, Fatemeh Mehrafrooz, Nele Mentens, Todor Stefanov. 2025 28th Euromicro Conference on Digital System Design (DSD), 2025, pp. 635-642. DOI: [10.1109/DSD67783.2025.00092](https://doi.org/10.1109/DSD67783.2025.00092).

# Abstract

Analog In-Memory Computing (AIMC) has emerged as a promising solution to address the performance and energy efficiency limitations of conventional von Neumann architectures for machine learning (ML) applications. This promising approach relies on analog-to-digital converters (ADCs) to enable the integration of AIMC tiles into larger digital systems. In this paper, we investigate the vulnerability of AIMC tiles to power side-channel attacks targeting these ADCs. Specifically, we demonstrate that the numerical values of weights stored in an AIMC tile, that are often a critical asset of an ML model, can be extracted by analyzing the power consumption of the ADCs. With this objective in mind, we propose TraceFormer, which is a two-phase method: 1) we train a Transformer neural network (NN) model to translate captured ADC power traces into digital output values; 2) we utilize a novel input-controlled weight isolation technique in order to isolate each individual weight within the AIMC tile, and then reveal the isolated weight’s value by ADC power side-channel analysis using the Transformer NN model. We demonstrate the practical applicability and robustness of our proposed method by power side-channel analysis of Oscillator-based ADCs that are typically integrated within AIMC tiles. The experimental results show high accuracy and robustness of our Transformer-based analysis, implying potential vulnerabilities in AIMC tiles.

# Key result

High extraction accuracy and robustness on oscillator-based ADCs emulated on real hardware, exposing a new attack surface in AIMC tiles.

# Cite

R. Siyadatzadeh, F. Mehrafrooz, N. Mentens and T. Stefanov, "TraceFormer: A Transformer-Based Method for Weight Extraction from AIMC Tiles," in 2025 28th Euromicro Conference on Digital System Design (DSD), Salerno, Italy, 2025, pp. 635-642, doi: 10.1109/DSD67783.2025.00092.

```bibtex
@inproceedings{siyadatzadeh2025traceformer,
  author    = {Siyadatzadeh, Roozbeh and Mehrafrooz, Fatemeh and Mentens, Nele and Stefanov, Todor},
  title     = {{TraceFormer}: A Transformer-Based Method for Weight Extraction from {AIMC} Tiles},
  booktitle = {2025 28th Euromicro Conference on Digital System Design (DSD)},
  year      = {2025},
  pages     = {635--642},
  publisher = {IEEE},
  address   = {Salerno, Italy},
  doi       = {10.1109/DSD67783.2025.00092},
  url       = {https://doi.org/10.1109/DSD67783.2025.00092},
}
```

# Files

* [Web page](https://roozbeh.nl/publications/traceformer-weight-extraction-aimc/)
* [Author manuscript (PDF)](https://roozbeh.nl/papers/TraceFormer.pdf)
* [AI-generated audio overview (NotebookLM), 4 min](https://roozbeh.nl/audio/TraceFormer.m4a)
* [AI-generated video overview (NotebookLM), 8 min](https://roozbeh.nl/videos/TraceFormer.mp4)

# Related

* [NeuroSoC: security assessment of an in-memory computing SoC](../projects/neurosoc.md)
* [Security Challenges in Analog In-Memory Computing](../notes/imc-security.md)
* [From Power Traces to Model Weights: P2W and TraceFormer](../notes/weight-extraction.md)
* [Phase-Change Memory for Analog In-Memory Computing](../notes/pcm-future.md)

Author record: [Roozbeh Siyadatzadeh](../person/roozbeh-siyadatzadeh.md).