Securing Non-Volatile Memory Against Physical and Microarchitectural Attacks

17 concepts8 papers1 proposal

For decades, hardware security models relied on a fundamental physical assumption: when a system loses power or deallocates memory, its secrets vanish. This ephemeral nature of volatile memory (DRAM) served as a natural defense mechanism. If an attacker physically walked away with a powered-down device, the cryptographic keys and sensitive user data stored in its RAM would quickly fade into electrical noise.

Modern computing architectures are rapidly dismantling this assumption. To overcome the energy walls and performance bottlenecks of traditional memory, the industry is shifting toward Non-Volatile Memory (NVM) technologies—such as STT-MRAM, FeFETs, and phase-change memory—alongside highly persistent volatile architectures. While these innovations offer massive gains in energy efficiency and crash consistency, they introduce a critical vulnerability: data outlives the execution context that created it.

When data persists indefinitely on a physical medium, physical possession of a device equates to full data access. Securing this new paradigm requires rethinking the boundary between storage, memory, and physical security.


The Threat of Physical Persistence

The vulnerability of data remanence is not entirely new, but its severity scales dramatically with memory persistence. In traditional DRAM systems, data does not disappear the exact millisecond power is cut; charges leak from capacitors gradually.

As explored by Yongjun Lee et al. in DRAM architecture for efficient data lifetime management, this window of vulnerability allows for "cold-boot attacks." An attacker with physical access can spray a liquid coolant onto a DRAM module to slow down charge leakage, physically extract the stick, and insert it into a malicious machine to read residual secrets. While software-based memory zeroing can mitigate this, it incurs a steep performance penalty (often a 5–11% slowdown). To solve this, the authors proposed a hardware-level modification that uses a simple "live flag" for DRAM rows to instantly isolate and clear deallocated data with virtually zero performance overhead.

Volatile DRAM (With Remanence)
[Power Loss] ──> [Slow Charge Leakage] ──> (Cold-Boot Attack Window) ──> [Data Lost]

True Non-Volatile Memory (NVM)
[Power Loss] ──> [Data Retained Indefinitely] ──────────────────────────> [Data Vulnerable Forever]

When we transition from DRAM to true Non-Volatile Memory, the attack window expands from a few minutes to infinity. Because NVMs retain their state without power, a stolen device exposes every piece of unencrypted data directly to physical extraction.

This vulnerability severely undermines modern security primitives like Trusted Execution Environments (TEEs). As highlighted by Jiamin Shen et al. in T-Edge: Trusted Heterogeneous Edge Computing, if a TEE secures on-chip execution but relies on unencrypted external RAM, physical access allows attackers to bypass the CPU's cryptographic boundaries entirely.


The Performance-Security Tax of Encryption

The intuitive solution to physical data extraction is straightforward: encrypt everything before it leaves the CPU cache. However, implementing robust memory encryption introduces a severe "security tax" that conflicts with the primary benefits of NVMs: speed and longevity.

Standard memory encryption relies on counter-mode block ciphers, which require maintaining encryption state counters. This introduces two major challenges:

  1. Latency: Fetching encryption counters adds extra memory round-trips, degrading performance.
  2. Write Amplification and Endurance: NVMs have limited write endurance. Frequently updating encryption counters and writing encrypted blocks accelerates the physical degradation of the memory cells.

To balance these trade-offs, Pengfei Zuo et al. introduced SecPM in SecPM: a Secure and Persistent Memory System for Non-volatile Memory. SecPM decouples crash consistency from security overhead by utilizing a counter cache write-through scheme alongside a counter write reduction strategy, minimizing the performance and wear-leveling penalties traditionally imposed by persistent encryption.

Taking a different architectural approach, Yixin Xu et al. proposed moving encryption directly into the memory array itself. In Embedding security into ferroelectric FET array via in situ memory operation, the authors demonstrated that ferroelectric FET (FeFET) arrays can leverage their inherent physical and operational characteristics to perform in situ encryption and decryption. By eliminating the need for bulky, power-hungry external AES engines, this method achieves a 90% reduction in latency, demonstrating that the physics of the memory medium can be harnessed as a defensive asset rather than a liability.


New Side-Channels and Microarchitectural Leaks

Physical data theft is not the only risk. The unique physical properties and microarchitectural optimizations required to make NVMs viable create entirely new side-channel attack surfaces.

Power and Magnetic Side Channels

Because NVMs rely on physical state changes (like changing magnetic orientation or ferroelectric polarization) rather than storing electrical charge, their write operations consume distinct amounts of power. Abhishek Chakraborty et al. exploited this in Correlation Power Analysis Attack against STT-MRAM Based Cyptosystems. By modeling the specific power consumption characteristics of Spin-Transfer Torque MRAM (STT-MRAM) during write cycles, the authors demonstrated a Correlation Power Analysis (CPA) attack capable of bypassing state-of-the-art countermeasures, proving that emerging NVMs introduce distinct power signatures that traditional cryptosystems fail to mask.

Furthermore, technologies like STT-MRAM are highly sensitive to external magnetic fields, making them vulnerable to disruption or denial-of-service attacks. To counter these multi-modal threats, Nikhil Rangarajan et al. developed SMART in SMART: A Secure Magnetoelectric AntifeRromagnet-Based Tamper-Proof Non-Volatile Memory. By utilizing magnetoelectric antiferromagnetic materials, the authors designed a memory architecture inherently resilient against magnetic tampering, power analysis, and photonic side-channel attacks.

NVM Physical Properties ──> Distinct Write Power Signatures ──> Correlation Power Analysis (CPA)
NVM Microarchitecture   ──> Hidden On-DIMM Buffers/Caches   ──> Microarchitectural Contention (NVLeak)

Microarchitectural Contention

Beyond the underlying physics, the commercial implementation of NVM systems introduces hidden complexity. Because NVM writes are slower than DRAM, manufacturers embed undocumented, off-chip microarchitectural components directly onto the memory modules—such as on-DIMM caches (NVCaches) and wear-leveling controllers—to mask latency and manage cell endurance.

As documented by Pradip Bose et al. in Energy-Secure System Architectures (ESSA): A Workshop Report, these efficiency-driven hardware optimizations frequently introduce unresolved security risks, including disturbance-based vulnerabilities akin to row-hammering variants.

These hidden layers can be actively exploited to exfiltrate data without physical tampering. In NVLeak: Off-Chip Side-Channel Attacks via Non-Volatile Memory Systems, Zixuan Wang et al. reverse-engineered the internal L2 NVCache of Intel Optane NVRAM systems. They discovered that an attacker can monitor contention on these off-chip caches to construct a high-bandwidth covert channel (up to 11 kbps). This attack successfully bypasses all CPU-centric defenses because the leakage occurs entirely on the off-chip memory module, outside the purview of traditional operating system or processor-level isolation.


Conclusion

The evolution of memory design from volatile to persistent fundamentally shifts the hardware security landscape. When memory behaves like storage, defenses can no longer assume that data boundaries disappear at power-off or deallocation. As persistent memory systems become standard across edge devices and data centers, securing them requires a holistic approach: one that accounts for the physical persistence of cells, the performance overhead of encryption, and the hidden microarchitectural side channels embedded within the silicon itself.

Key Foundational Work

Correlation Power Analysis Attack against STT-MRAM Based Cyptosystems (2017) — contributed 1 concept elements to this direction.
SecPM: a Secure and Persistent Memory System for Non-volatile Memory (2018) — contributed 1 concept elements to this direction.
DRAM architecture for efficient data lifetime management (2017) — contributed 1 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_cold_boot_attack_vulnerability_6c3c7467,
  title = {cold-boot attack vulnerability},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/cold-boot-attack-vulnerability/rdr_6c3c74675a4b47d583f30e74ede53e83},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_cold_boot_attack_vulnerability_6c3c7467_v1,
  title = {cold-boot attack vulnerability},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/cold-boot-attack-vulnerability/rdr_6c3c74675a4b47d583f30e74ede53e83/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake