TL;DR
This paper introduces a novel, formally secure approach to split manufacturing for protecting integrated circuit designs. By "locking" the front-end-of-line (FEOL) with a secret key embedded in the back-end-of-line (BEOL), they make designs indecipherable to untrusted foundries, overcoming limitations of previous methods. They also propose a "poor man's split manufacturing" using package-level routing for a more practical and cost-effective solution.
What Problem Does This Paper Address?
The semiconductor industry increasingly relies on outsourced fabrication, which introduces significant security risks. These include intellectual property (IP) piracy, overbuilding of designs, and the insertion of hardware Trojans. Split manufacturing, a technique where a chip design is divided into two parts (FEOL and BEOL) fabricated at different facilities, was proposed to mitigate these threats. However, existing split manufacturing methods are vulnerable to "proximity attacks." These attacks exploit the deterministic nature of standard Electronic Design Automation (EDA) tools to infer missing connections in the FEOL layout, thereby compromising the design's security. This paper addresses the need for a formally secure split manufacturing paradigm that is resilient to these attacks and also considers practical implementation challenges.
Methods
The core of the proposed solution is a novel paradigm: "lock the FEOL, unlock at the BEOL." This approach adapts the concept of logic locking. Instead of directly protecting the entire design, they focus on securing the FEOL layout by embedding secret keys. These keys are then implemented through the BEOL in a way that makes them indecipherable to foundry-based attackers.
Key techniques include:
- Logic Locking Adaptation: Secret keys are introduced via "key-gates" within the FEOL. These key-gates are driven by "TIE cells" (cells providing constant logic 0/1 signals).
- FEOL Security: To prevent proximity attacks, the placement of TIE cells is randomized. This disrupts any physical proximity hints that could reveal key-gate connections.
- BEOL Key-Net Routing: Key-nets (nets connecting TIE cells to key-gates) are "lifted" to the BEOL. This means they are routed in higher metal layers, making them inaccessible to an FEOL-centric attacker.
- "Poor Man's Split Manufacturing": For increased practicality and reduced cost, an alternative strategy is proposed. Instead of splitting at the BEOL, key-nets are routed to I/O ports and then to the package level. This bypasses the need for a dedicated, trusted BEOL foundry, relying instead on trusted packaging or board-level assembly.
Datasets
The paper evaluates its proposed scheme using:
- ITC'99 benchmarks: These are large-scale integrated circuit designs used for performance and security evaluations.
- ISCAS benchmarks: These are smaller benchmarks used for comparison with prior art.
The authors also release their framework and benchmarks publicly.
Key Insights and Ideas
- Formally Secure Split Manufacturing: The paper establishes a formal security definition for split manufacturing and proves that their proposed scheme meets this definition, offering a significant advancement over heuristic-based approaches.
- Key-Net Focus: The security of the proposed scheme relies solely on the protection of key-nets, not all nets. This simplifies the problem and allows for more efficient solutions.
- "Poor Man's Split Manufacturing" Viability: The idea of using package-level routing for IP protection is a practical and cost-effective alternative to traditional split manufacturing, especially for designs where a dedicated BEOL foundry is not feasible.
- Security Independence from Split Layer: The security of the key-nets is shown to be largely independent of the specific metal layer chosen for splitting, offering flexibility in design.
- Attack Resilience: The proposed scheme is demonstrated to be secure against state-of-the-art proximity attacks, with attackers performing no better than random guessing for key-nets.
Key Results
- Security Against Proximity Attacks: The logical CCR (Correct Connection Rate) for key-nets is consistently around 49%, indicating that attackers cannot reliably infer the correct key-bits. Physical CCR for key-nets is near 0%, demonstrating the effectiveness of the lifting technique.
- Low Overhead: For large-scale ITC'99 benchmarks, the proposed scheme incurs minimal overhead in terms of area (~4%) and performance (~0.5% to 0.7%). Power overhead is slightly higher but still acceptable.
- "Poor Man's Split Manufacturing" Benefits: This approach requires only half the number of additional signal bumps compared to protecting regular nets, leading to reduced die area and cost. It also allows for more power/ground bumps and relaxed bump spacing.
- Comparison to Prior Art: The proposed scheme is shown to be competitive or superior to existing methods in terms of protection metrics (PNR, CCR, HD, OER) on ISCAS benchmarks.
Limitations and Future Work
- Formalism for Combinational Circuits: The formal security analysis is presented for combinational circuits, though the authors note it can be extended to sequential designs.
- Potential for Invasive Attacks: While the scheme is secure against foundry-based proximity attacks, it does not directly address invasive attacks on fabricated chips (e.g., optical probing) or mask manipulation. The authors suggest that countermeasures for these can be applied on top of their scheme.
- Empirical vs. Formal Security: While the scheme is formally proven secure, the empirical validation relies on known attacks. The authors acknowledge that future attacks might emerge, but their framework can be extended.
- "Poor Man's Split Manufacturing" Trade-offs: While more practical, this approach might be perceived as weaker than traditional split manufacturing in terms of the number of hidden nets. However, its security is derived from hiding the key-nets, which is sufficient for provable security.
Future work could explore extending the formal analysis to sequential circuits, investigating countermeasures against invasive attacks, and further optimizing the "poor man's split manufacturing" approach for even greater cost-effectiveness and security.
Summary updated