Neutralizing Spatial Hints in Split Manufacturing

28 concepts11 papers1 proposal

The globalization of the semiconductor supply chain has led to a "fabless" business model where hardware designers rely on external, untrusted foundries to fabricate their integrated circuits (ICs). To prevent intellectual property (IP) theft and malicious tampering, designers use split manufacturing. As outlined by Johann Knechtel et al., this defense splits the physical layout into two parts: the lower layers (Front-End-of-Line, or FEOL) are fabricated by an untrusted foundry, while the upper metal layers (Back-End-of-Line, or BEOL) are manufactured at a trusted facility.

The security of split manufacturing relies on the assumption that the untrusted foundry cannot guess the missing global connections. However, a fundamental optimization loop in modern Electronic Design Automation (EDA) tools inadvertently undermines this assumption, giving rise to highly accurate proximity attacks.

The Proximity Vulnerability: How Optimization Leaks Topology

Why can an attacker reconstruct a missing netlist simply by looking at a partial physical layout? The answer lies in how EDA tools operate.

Standard physical design tools are hardwired to optimize for Power, Performance, and Area (PPA). To minimize signal propagation delay and power consumption, placement and routing algorithms minimize total wire length. Consequently, the tools naturally place logically connected components in close physical proximity.

This optimization creates a massive side channel: spatial hints. An untrusted foundry looking at the FEOL layout does not just see isolated gates; they see a spatial map of functional intent. Because physical distance strongly correlates with logical connectivity, an adversary can use spatial heuristics to statistically guess the hidden BEOL connections.

[Logical Netlist] ---> [EDA Tool: Minimize Wirelength] ---> [Physical Layout]
                                                                   |
                                                      (Leaks Spatial Hints)
                                                                   v
[Missing Connections] <--- [Proximity Attack Heuristics] <---------+

As Satwik Patnaik et al. note, this vulnerability persists even if individual gates are camouflaged, because the spatial arrangement of the cells still betrays the underlying circuit topology. To make matters worse, attackers do not have to rely on manual guessing. Haocheng Li et al. demonstrated that deep learning models can combine numerical design data with image-based layout snapshots to systematically exploit these proximity hints, reconstructing missing layers with high precision in a fraction of the time required by traditional attacks.

(Note: This relationship between physical layout and information leakage is a universal challenge in hardware security. Similar spatial vulnerabilities appear when deep learning models exploit physical input spacing to infer keystrokes on point-of-sale keypads, as shown by Stefano Cecconello et al., or when memory layout boundaries affect side-channel resistance, as explored by Christos Andrikos et al..)

To secure split manufacturing, designers must break this correlation. They must neutralize spatial hints so that physical proximity no longer equals logical connectivity.

Weaponizing the EDA Tool: Netlist Randomization

If proximity attacks succeed because EDA tools optimize for the true netlist, one elegant defense is to trick the tools into optimizing for a false one.

Satwik Patnaik et al. proposed a technique that randomizes the netlist before it undergoes physical placement and routing. By introducing dummy connections or altering the logical description fed into the layout software, the automated tools are forced to optimize for incorrect signal paths.

Because the EDA tool blindly minimizes wire length for this misleading netlist, it clusters components that should not be together and separates components that should be close. When the untrusted foundry attempts to run a proximity attack on the resulting physical layout, they are fed entirely misleading spatial hints, driving their connection success rate down to near zero percent. The true functionality is only restored later during the trusted BEOL manufacturing phase.

Disrupting Short-Wire Heuristics: Wire Lifting and Placement Constraints

Instead of changing the netlist, other defenses directly disrupt the short-wire heuristics of physical design tools by forcing physical separation.

Concerted Wire Lifting

Satwik Patnaik et al. introduced Concerted Wire Lifting, a method that uses custom "elevating cells" to selectively move critical interconnections to higher, trusted metal layers. By precisely controlling the physical distance between the resulting "open pins" (Satwik Patnaik et al.), the layout engine increases the spatial distance between connected wire segments. This intentional pin displacement removes the localized spatial hints that attackers rely on to resolve missing wiring.

Standard Layout:
[Gate A] --(Short Wire)-- [Gate B]  <-- High Proximity Hint

Wire-Lifted Layout:
[Gate A] --(Pin Displaced)-------------> [Elevating Cell] --(Trusted Layer)--> [Gate B]

Placement-Centric Constraints

Rather than randomizing the entire layout—which incurs heavy performance penalties—designers can inject specific geometric constraints into the CAD flow. Abhrajit Sengupta et al. demonstrated that forcing connected components into distinct physical regions, or clustering gates by alternative logical functions, breaks down network-flow attacks. This approach allows designers to strategically increase wire length and unpredictability only where necessary, bounding the overhead cost to power and delay.

Neutralizing Key-Gate Proximity

This concept is equally vital when split manufacturing is combined with logic locking. If an attacker can look at the layout and see exactly which key-gates are close to specific static logic drivers (TIEHI/TIELO cells), they can deduce the secret key. To prevent this, Abhrajit Sengupta et al. and Abhrajit Sengupta et al. developed techniques to distribute key-providing cells randomly across the chip layout and route their connections to higher layers. This breaks the deterministic proximity correlations, forcing attackers at untrusted foundries to resort to random guessing.

Moving to the Third Dimension: 3D IC Routing Obfuscation

The emergence of 3D Integrated Circuits (3D ICs) provides a structural architectural solution to the proximity problem. Instead of trying to hide connections on a flat 2D plane, designers can split the manufacturing across vertically stacked physical tiers.

Satwik Patnaik et al. introduced a security-driven CAD flow that integrates split manufacturing with 3D stacking. By decoupling the physical placement of logic gates on individual tiers from their vertical connections, the design's functional intent remains hidden.

The core mechanism relies on randomizing the routing of the vertical interconnects (vias) between the stacked tiers. As explained by Satwik Patnaik et al., vertical connections are routed through a redistribution layer that uses randomized via placement and hardware switchboxes.

Tier 2 (Trusted)   :  [Gate B]             [Gate C]
                         |                    |
Randomized Vias    :   /                      \    <-- Breaks Vertical Proximity
                      /                        \
Tier 1 (Untrusted) :  [Gate A]             [Gate D]

Because of this vertical randomization, two components that sit directly on top of each other on adjacent tiers are not necessarily connected. This completely eliminates the spatial correlation between components across tiers, rendering proximity-based reverse engineering mathematically impractical while maintaining highly efficient, low-overhead local routing within the individual tiers.

Key Foundational Work

Raise your game for split manufacturing: restoring the true functionality through BEOL (2018) — contributed 4 concept elements to this direction.
Concerted wire lifting: Enabling secure and cost-effective split manufacturing (2018) — contributed 3 concept elements to this direction.
Best of Both Worlds: Integration of Split Manufacturing and Camouflaging into a Security-Driven CAD Flow for 3D ICs (2018) — contributed 3 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_proximity_based_netlist_reconstruction_00a64d37,
  title = {proximity-based netlist reconstruction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/proximity-based-netlist-reconstruction/rdr_00a64d3799f84fe6a40e52819d92574d},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_proximity_based_netlist_reconstruction_00a64d37_v1,
  title = {proximity-based netlist reconstruction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/proximity-based-netlist-reconstruction/rdr_00a64d3799f84fe6a40e52819d92574d/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake