This research paper solves an important, yet engineering, problem: it protects private keys from leakage via REE and makes support for new chains scalable. A quantum threat is a mathematical problem orthogonal to the PDF threat model. But it is precisely the modularity of the architecture that makes it a valuable candidate for a migration path to post-quantum cryptography: if a signature algorithm is encapsulated in a TA with a VFT interface, its replacement is an OTA update, not a device reflash.
Content
- Two crises, one question
- Mechanism: Why ECDSA Breaks and TEE Doesn’t
- What exactly does a modular TEE protect and what does it not protect?
- Analysis of 2506.17988v1: Architecture as Transport
- Bitcoin Specifics: When a Public Key is Actually Available
- Resource model: logical and physical qubits
- Migrating to PQC via Modular TA: A Testable Scenario
- Experimental Plan and Metrics
- Limitations, fair comparison, risks
- Practical value and economics
- Open questions and the next step
- Conclusion
- Sources
1. Two crises, one question
How is it possible that a wallet protected by a hardware TEE would be useless against an attack that has not yet been implemented—but has already been described mathematically?
The work of Kim, Yun, and Jang begins with sobering statistics: according to Chainalysis, approximately $2.2 billion in digital assets will be stolen in 2024, with compromised private keys accounting for 43.8% of all stolen funds—more than smart contract bugs and oracle manipulation combined. The five largest exchange hacks (Bybit in 2025, DMM Bitcoin in 2024, FTX in 2022, Coincheck in 2018, and Mt. Gox in 2014) collectively cost the industry billions of dollars, and in each case, a key leak or weak key management was the key factor.
The authors’ solution is architectural: move all private key operations to ARM TrustZone / OP-TEE , break the functionality into modular Trusted Applications (TA) , one per blockchain, deliver them via a signed FOTA , and verify them with a two-step signature (developer → registry). This closes an entire class of attacks across the six layers of the threat model (memory, OS, software, network, protocol, etc.).
But there’s a second crisis, only briefly mentioned in the PDF—in line e-3, “Cryptographic Breaks (Weak RNG/Key Reuse)” and in the phrase “or had a quantum computer.” The point is that the cryptography that underlies these keys is mathematically vulnerable to Shor’s algorithm . And this raises an interesting paradox:
The TEE protects the key from theft now . But if the signature algorithm executed within the TEE is later broken by a quantum computer , the TEE will honestly and securely sign the transaction, resulting in the loss of funds.
Hence, the central test question of the paper: can the modular TEE architecture of 2506.17988v1 be not only a “key safe”, but also a vehicle for replacing a cryptographic algorithm—that is, for migrating to post-quantum cryptography (PQC) without changing the device?
This is precisely the connection suggested by Section 9 of the Research Constructor: questions 49–56 on cryptography, cryptanalysis, and security, including question 54, “For the Bitcoin topic: which mechanism is being investigated—signatures, hashing, or protocol—and when is the required public key available to an attacker?” We’ll see that the answer to this question changes the entire risk assessment.
1.1 Testable hypothesis
H1. The modular TEE architecture with OTA TA delivery and two-stage signature allows replacing the cryptographic signature module (e.g., ECDSA/secp256k1 with ML-DSA or a hybrid scheme) without changing the base firmware and without changing the user device , maintaining the integrity of the supply chain.
Refutation criterion: if changing the signature scheme requires rebuilding the entire TEE OS or reflashing the bootloader, and not just replacing the TA module via FOTA, then H1 is refuted in its strict form and reduces to the weaker statement “TEE is more convenient than nothing.”
1.2 What is NOT stated
- There is no claim that a quantum computer capable of breaking secp256k1 exists today.
- No specific date has been set for the “quantum deadline”.
- It is not claimed that TEE makes PQC migration free – only that it makes it cheaper and more secure across the supply chain .
2. Mechanism: Why ECDSA Breaks and TEE Doesn’t
For an article to be scientific rather than journalistic, two levels must be distinguished: the mathematical robustness of the algorithm and the engineering isolation of the computation . TEE enhances the latter and has no effect on the former.
2.1. Mathematics secp256k1 and ECDSA
Bitcoin (and Ethereum) use the secp256k1 curve:
y² ≡ x³ + 7 (mod p), p = 2²⁵⁶ − 2³² − 977
The private key d is a random 256-bit number. The public key is the point P = dG, where G is the generator. The ECDSA signature for the message hash m is:
k ∈ [1, n−1] random; R = kG;
r = x_R mod n; s = k⁻¹ (H(m) + d·r) mod n
The security relies on the elliptic curve discrete logarithm problem (ECDLP) : given G and P = dG, find d. The classically best known algorithm—Pollard’s ρ—requires ~√n ≈ 2¹²⁸ operations. This is computationally infeasible on classical machines.
2.2 What does Shor’s algorithm do?
Shor (1994) showed that ECDLP can be solved in polynomial time on a quantum computer with a quantum Fourier transform. This isn’t a heuristic or a “squared speedup”—it’s a change in complexity class. It’s this fact that makes secp256k1/ECDSA fundamentally vulnerable, not just “slowly breakable.”
For symmetric cryptography, the picture is milder: Grover’s algorithm only provides a quadratic speedup, meaning SHA-256 maintains ~128-bit resistance to preimage attacks. Therefore, a quantum threat to blockchain is primarily a threat to signatures, not hash functions . This is an important clarification that is often lost in popular texts.
2.3. What does TrustZone do?
ARM TrustZone divides the system into the Normal World (Android) and the Secure World (OP-TEE). Memory and peripherals are marked as Secure/Non-Secure; access is only through an SMC instruction under the control of the Secure Monitor. The Normal World cannot physically read the Secure World. Within the Secure World, OP-TEE isolates the TAs from each other, and the GP Crypto API provides standardized access to the algorithms.
Key observation: TrustZone is about confidentiality and computational integrity, not algorithmic security . It doesn’t “slow down” Shor or “hide” the key to the quantum algorithm: if the public key P falls into the hands of an attacker with a quantum computer, the TEE is powerless—d is recovered from P mathematically, without the device’s intervention.
The distinction I propose to make as the thesis of this paper is that the modular TEE of 2506.17988v1 solves the key leakage problem (§1.1 PDF) and the trust problem in the module supply chain (§1.3 PDF), but it does not solve the problem of the algorithm’s mathematical vulnerability . The value of this architecture is that it can serve as a vehicle for solving the second problem—if the TA is designed so that the signature algorithm is replaceable.
3. What exactly does a modular TEE protect and what does it not protect?
The PDF contains Table 2, a comparison of attack vectors. It’s useful to expand it with a “quantum” line so the reader can immediately see the applicability limits.
| Attack vector | Monolithic wallet | Proposed protection in PDF | Quantum attack |
|---|---|---|---|
| REE malware (sniffing) | Key operations are open | All operations within TA | not applicable attack not through REE |
| OTA supply chain | Update injection is possible | Two-step signature + verification at the TEE-OS level | The PQC module also protects usefully |
| InterTA side-channel | General process → leak | TA isolation, cache separation, S(G)PAN | useful for changing algorithms |
| New chain | A vendor patch is needed | Community-driven modules | It’s really useful – the same mechanism for PQC |
| Weak RNG / nonce reuse | Key recovery is possible | CSPRNG inside TA (TA1) | partially eliminates one subclass |
| Quantum ECDLP (Shor) | Vulnerable mathematically | TEE is not eliminated | the algorithm needs to be changed |
This table is already a standalone contribution: it shows that the quantum threat is orthogonal to the PDF threat model. Of the six layers of Table 5 of the PDF, the quantum attack falls in layer (e) Blockchain Protocol , item e-3 Cryptographic Breaks —and this is the only layer where the TEE architecture alone does not provide a response.
4. Analysis of 2506.17988v1: Architecture as Transport
Now let’s look at PDF not as “just another wallet,” but as a mechanism for delivering code to a secure environment . This is precisely what makes it a candidate for the role of PQC migration transport.
4.1 The three components and their roles
- Wallet Application (CA) in REE – UI and network; does not have access to keys.
- Secure World has one TA module per blockchain: BTC Module, ETH Module, and Unspecified Module. Inside are TA1 (Key Generation) and TA2 (Signing), plus additional TA modules as needed.
- Module Update Server — CI/CD pipeline: static analysis → symbolic execution → fuzzing; then signing and publishing to the version registry and CDN.
4.2. Six steps of the module life cycle
- The developer compiles the module for “FooChain”, signs it with GPG/PKCS#11, and uploads it.
- The server automatically validates; policy violations are rejected.
- Approved packages are published and the manifest is updated.
- User tags FooChain → CA pulls OTA and streams to Secure World.
- TEE-OS verifies RSA-PSS and SHA-256, maps TA into an independent address space.
- When a new version is released, delta-OTA and hot-swap will be available after the same testing.
Steps 5–6 are the actual algorithm replacement process . If the TA module encapsulates “chain-specific logic such as key generation, address derivation, signing, and handling of BIP-32/44 parameters,” then replacing ECDSA with ML-DSA is a matter of replacing one TA module in the set, not rebuilding the entire firmware.
4.3. What does PDF say about the future?
Section 6 (Future work) already contains templates that can be reused for PQC:
- §6.3 — CI/CD with a role-based model (admin / validator / contributor) and transparent logging (Sigstore rekor);
- §6.4 — Wallet Manager HAL, SELinux domains tee_exec, fota_update, wallet_service; integration with Verified Boot and dm-verify;
- §6.5 — TA SDK, virtual function table (VFT) instead of #ifdef for secp256k1/ed25519/sr25519;
- §6.7 — comparison with ARM CCA and Intel TDX, formal verification via Tamarin/Coq.
The point about VFT is key. It literally describes the interface through which ML-DSA or a hybrid scheme can be inserted without changing the rest of the code.
4.4. The Weakness of PDF from the Quantum Point of View
The PDF doesn’t discuss post-quantum migration as a separate goal anywhere . The threat model only mentions “quantum computer” in line e-3 and doesn’t elaborate on the resource estimate. This isn’t a flaw in the work—it’s a gap that this article fills: the PDF provides a transport, but doesn’t provide a target algorithm or its acceptance criteria.
5. Bitcoin Specifics: When the Public Key is Actually Available
This is perhaps the most interesting and most verifiable point. Question 54 of the constructor asks, “When is the required public key available to an attacker?” The answer radically changes the risk assessment—and that’s a good hook for the reader.
5.1. P2PKH: the key is hidden until the first spend
A standard Bitcoin address is HASH160(P) = RIPEMD160(SHA256(P)), encoded in Base58Check. Only the hash of the public key is stored on the blockchain. The public key P is only added to the blockchain at the time of spending —in the scriptSig of the input transaction.
Consequence: a quantum attacker with Shor’s algorithm can attack:
- Addresses from which funds have already been spent (the public key is permanently stored in the blockchain);
- Reused addresses (public key already exposed by previous spending);
- Transactions in the mempool are the window between broadcast and confirmation, when P is visible but funds have not yet been moved. The attack speed relative to block time (~10 minutes for Bitcoin) is critical here.
This echoes PDF §2.1.4: ” Identities on the blockchain are represented by addresses derived from public keys… yet every transaction is still linkable to a pseudonymous address .” Pseudonymity is not the same as cryptographic secrecy of a key.
5.2. Other PDF diagrams – varying degrees of exposure
| Net | Key / curve | Signature | Address | When P becomes available |
|---|---|---|---|---|
| Bitcoin | secp256k1 | ECDSA | Base58Check (P2PKH) = RIPEMD160(SHA256(P)) | On first spend; in the mempool – a short window |
| Ethereum | secp256k1 | ECDSA (+ recovery ID) | Hex, low 20 bytes of Keccak256(P) | On the first outgoing transaction; in the mempool – window |
| Algorand | Curve25519 / Ed25519 | EdDSA | Base32 (public key + 4B checksum) | Almost always , the address is the public key. |
| Polkadot | Curve25519 | Schnorrkel / Sr25519 | SS58 (Base58 public key + prefix + checksum) | Almost always – like Algorand |
This is an important conclusion that is rarely mentioned in popular texts: the quantum risk for Algorand and Polkadot is higher than for Bitcoin P2PKH because their address formats expose the public key continuously, not just at the time of spending. For Bitcoin, the risk is concentrated in reused addresses and the mempool window. For Ethereum, it is in reused accounts and the mempool window.
This is exactly the type of “verifiable statement” required by the constructor (questions 49–54): not “Bitcoin will be hacked,” but “under what conditions and at what point does the public key of a particular scheme become available to an attacker.” This is precisely the formulation that makes the article scientific.
6. Resource model: logical and physical qubits
To avoid sensationalism, it is necessary to clearly distinguish between three quantities that are often confused:
- logical qubits are abstract qubits of Shor’s algorithm;
- physical qubits – real qubits with noise that require correction;
- execution time – taking into account the clock frequency and correction codes.
6.1. Estimates from the literature (ranges, not points)
For 256-bit ECDLP, different groups obtain orders of magnitude similar results: approximately 1.5–2.5 thousand logical qubits and approximately 10⁹–10¹⁰ quantum gates (Toffoli-like) per attack. Several orders of magnitude more physical qubits are required: with realistic error levels (~10⁻³) and a shallow code, millions of physical qubits are required, with execution times ranging from hours to days. More advanced codes and architectures lower the estimate but do not change the order of magnitude.
Important: These are estimates from literature , not measured facts. They should be cited with the error model, correction architecture, and clock rate assumptions—otherwise, they become mere marketing figures.
6.2 Why is this important for wallet design?
The resource model determines the migration window . If the “millions of physical qubits” estimate is correct, the industry has years, not months. However, if a breakthrough in error correction occurs, the window will shrink. This leads to a practical requirement: the wallet architecture must allow the algorithm to be replaced faster than the estimate changes . This requirement is the criterion for testing H1.
6.3. Comparison with Grover
For symmetric primitives, the quantum threat is weaker: Grover yields √N, meaning SHA-256 maintains ~128-bit security. This means Bitcoin P2PKH hashes remain relatively secure even in the quantum era—but only until the public key is revealed. Once P is revealed, the security switches from hashing to ECDLP, and this is where Shor comes in.
7. Migration to PQC via Modular TA: A Testable Scenario
Now let’s put it all together. Below is a scenario that can be implemented on the PDF architecture and that provides testable points of failure.
7.1 Target standards
On August 13, 2024, NIST approved FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), and FIPS 205 (SLH-DSA). ML-DSA (Dilithium) and, to a lesser extent, SLH-DSA (SPHINCS+) are relevant for blockchain transaction signatures—they are not a one-to-one replacement for ECDSA, but they provide post-quantum security. NCCoE is leading a migration project—practical context is missing from the PDF.
7.2. Three migration options
| Option | What’s changing? | Compatibility | The role of modular TEE |
|---|---|---|---|
| A. Complete replacement | ECDSA → ML-DSA in TA2; new address format | Requires protocol/wallet level support | Replacing one TA module via FOTA |
| B. Hybrid scheme | ECDSA + ML-DSA simultaneously; check both | Backward compatibility, but more data per transaction | Additional TA in the set; VFT from §6.5 PDF |
| C. Migration for new addresses only | The old P2PKHs remain, the new ones are PQCs | Gradually, but requires discipline from users | New module + selection rule in CA |
7.3. What does the PDF architecture provide in each variant?
- Isolation: The PQC module does not see keys from other chains (least privilege principle from §5.5 PDF).
- Delivery: OTA with two-step signature – the same mechanism as for FooChain.
- Rollback: Verified Boot + rollback index from §6.3 PDF protects against substitution with an older vulnerable version.
- Audit: Houy et al. (2023) explicitly point to the need for dynamic architectures; the modular TEE addresses this need.
- Performance: ML-DSA signatures are larger than ECDSA; PDF §6.5 sets a target of “<1 ms for 256-bit ECDSA” – for PQC, the target will have to be revised, and this is a measurable task.
7.4 Where H1 can break down
- Keys and addresses: ML-DSA produces a public key of ~1.3 KB and a signature of ~2.4 KB, compared to 33/64 bytes for secp256k1. This changes the transaction format, and therefore the protocol, not just the TA. H1 strictly requires the protocol to support the new format.
- Secure Storage: PDF §5.5 discusses secure storage in TEEs. PQC keys are larger; it’s important to check whether the allocated TA memory is sufficient (PDF §6.5 mentions a 1 MB heap limit).
- Performance: ML-DSA-44 verify/sign on Cortex-A78 needs to be measured; the <1 ms target from the PDF for ECDSA is not automatically carried over.
- Attestation: PDF §2.3.4 notes that TrustZone does not have universal remote attestation. For PQC migration, this means that trust in the “correct” module version relies on the local chain (Verified Boot + registry signature) rather than remote attestation.
8. Experiment plan and metrics
To ensure this article is not merely conceptual, below is a concrete plan for verifying H1 on the PDF architecture. The metrics are taken from §6.6 of the PDF and expanded with PQC specifics.
8.1 Functional tests
- Assemble TA module with VFT interface (secp256k1 / ML-DSA-44 / hybrid).
- Run CA↔TA IPC via conn_test_app (as in §6.1.1 PDF) for three modes.
- Check that replacing the module via FOTA does not require rebuilding the TEE-OS.
- Check rollback: attempt to install an old version of the module should be blocked by rollback index.
8.2. Metrics (with goals)
| Metrics | Unit | Purpose (based on §6.6 PDF) | What does it check? |
|---|---|---|---|
| ECDSA signature latency | ms | ≤ 1.5 ± 0.2 | Baseline |
| ML-DSA-44 Signature Latency | ms | measure; compare with ECDSA | PQC price |
| TA module size | KB | fits within the secure storage limit | Limitation §6.5 PDF |
| OTA time 10 MB | With | < 15 sec @ Wi-Fi 6 | Transport §6.3 PDF |
| Peak memory TA | KB | ≤ 1 MB heap | Limitation §6.5 PDF |
| Hot-swap success | % | 100% after RSA-PSS + SHA-256 | Integrity |
| Rollback when downgrading | binary | is blocked | Rollback protection |
8.3. Security tests
- Fuzzing the PQC module’s transaction parser (AFL++ / libFuzzer, as in §6.6 PDF).
- Isolation check: TA PQC should not read keys of BTC/ETH modules.
- Verifying that the private key does not leave Secure World: There should be no API at the CA level that returns d.
- Two-step signature verification: Compromising the developer key without the registry signature should not allow installation.
8.4 What constitutes a refutation of H1
H1 is strongly disproved if at least one of the following conditions is met:
(a) changing the signature scheme requires changing the bootloader or TEE-OS;
(b) the size of the PQC key/signature does not fit within the secure storage and heap limits, and this is not addressed architecturally;
(c) on-device signature verification takes time that is incompatible with UX (the threshold is pre-set, e.g. > 500 ms);
(d) the two-step PDF signature mechanism §1.3.3 does not scale to PQC modules without changing the manifest format.
9. Limitations, fair comparison, risks
9.1 What the article does not prove
- That the PQC blockchain migration will be quick and painless.
- That a modular TEE is the only way; hardware wallets and HSMs can migrate too.
- That Shor’s resource estimates are final.
9.2 Competing explanations and approaches
| Approach | Pros | Cons | Comparison with TEE modules |
|---|---|---|---|
| Hardware wallet (Ledger/Trezor) | Insulation, a proven market | Single device, UX, supply chain | PDF §1.2: Still vulnerable to supply chain and phishing |
| Intel SGX | Small TCB, remote attestation | Not a mobile scenario; limited secure memory | Does not replace TrustZone on a smartphone |
| AMD SEV | Isolation of entire VMs | Not for mobile wallets | Another niche |
| ARM CCA | A stronger model than TrustZone | Maturity, accessibility | PDF §6.7 offers a comparison – an open question |
| Pure PQC without TEE | Easier | Keys in REE → leak | Does not resolve §1.1 PDF |
9.3. Dual-use and dependency risks
- Ledger centralization: The “verified contribution” model of PDF §1.3.6 requires trust in the ledger; it is a point of failure.
- Supply-chain: Two-step signature reduces, but does not eliminate, the risk of registry key compromise.
- Transparency: Sigstore rekor (PDF §6.3) provides auditing, but requires infrastructure.
- Vendor dependency: OP-TEE on a device is also dependent on the OEM firmware.
10. Practical value and economics
Who needs it and what benefit can be measured (designer questions 63–64):
- For the user: one installation – one wallet – multiple chains; algorithm migration without changing devices. Metric: percentage of successful OTA updates of the PQC module without access loss.
- For the auditor: one TA interface, not ten monolithic wallets. Metric: reduction of duplicate audits (PDF §1.2, point 3).
- For chain developer: community can contribute modules; metric: time from specification publication to working TA.
- Regulator: auditable supply chain and transparent log. Metric: share of modules with a complete signature chain.
10.1 Total Cost of Ownership
PDF doesn’t provide cost estimates, and that’s fair. For the purposes of this article, it’s enough to state: PQC signatures are larger, OTA packets are larger, and secure storage is more expensive. These are measurable tradeoffs, not “security for free.”
11. Open questions and the next step
For section 12 of the designer (questions 69–70):
- Formal Verification of the VFT Interface (PDF §6.7): Can We Prove That Modulus Replacement Does Not Violate the “Private Key Never Leaves the Secure World” Invariant? Tools: Tamarin, Coq.
- Remote attestation for TrustZone: PDF §2.3.4 notes the lack of a universal standard. This is critical for PQC migration: how does the remote party know that the device actually has a PQC module?
- Hybrid Schemes in Blockchain: How to Combine ECDSA and ML-DSA in a Single Transaction Without Inflating It to Unacceptable Size?
- ARM CCA vs. TrustZone: Does CCA Provide Stronger Guarantees for PQC Modules?
- UX Threshold: What PQC signature latency on a smartphone is considered acceptable? This is an empirical question, not a theoretical one.
- Cross-chain migration: If Bitcoin and Ethereum migrate at different rates, how does a modular wallet manage the hybrid state?
The following experiment, which best distinguishes between competing explanations, is to assemble a TA module with a VFT interface, perform a secp256k1 → ML-DSA-44 swap via FOTA, and measure (a) latency, (b) memory, and (c) isolation preservation . If all three metrics are within the PDF limits in §6.5–6.6, H1 is confirmed in engineering form. If not, we obtain a specific list of obstacles rather than general reasoning.
12. Conclusion
Work 2506.17988v1 solves an important, yet engineering, problem: it protects private keys from leakage via REE and makes support for new chains scalable. A quantum threat is a mathematical problem orthogonal to the PDF threat model. But it is precisely the modularity of the architecture that makes it a valuable candidate for a migration path to post-quantum cryptography: if a signature algorithm is encapsulated in a TA with a VFT interface, its replacement is an OTA update, not a device reflash.
The paper’s testable conclusion: H1 is confirmed in engineering form when three conditions are met : (1) the TA interface supports circuit replacement without rebuilding the TEE OS; (2) PQC keys and signatures are within the secure storage and heap limits; (3) verification/signature latency remains within the UX threshold. Each condition is measurable on the PDF architecture, and this is what turns the popular thesis “a quantum computer will break Bitcoin” into a research program.
The main takeaway for the reader: The quantum threat to blockchain isn’t a matter of “everything will collapse tomorrow” or “it doesn’t concern us.” It’s a question of the migration window . The architecture of 2506.17988v1 provides a mechanism that can shorten this window—but only if TA is designed from the ground up with algorithm replacement in mind. This article shows how to test this, not how to believe it.
Sources
- S. Kim, S. Yun, J. Jang. Secure User-Friendly Blockchain Modular Wallet Design Using Android & OP-TEE. arXiv:2506.17988v1, 2025. (Main source analyzed: §1.1 Incidents; §1.2 Motivation; §1.3 Architecture; §2.2 Signature Schemes; §2.3 TEE; §2.4 FOTA; §3 Open-TEE/SBLWT/Houy/TBW Overview; §4 Threat Model; §5 Design; §6 Future Work.)
- Chainalysis. Crypto Crime Report , 2024–2025 — statistics: $2.2 billion and 43.8% of private keys (quoted from PDF §1.2).
- PW Shor. Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer. SIAM J. Comput., 1997.
- LK Grover. A fast quantum mechanical algorithm for database search. STOC, 1996.
- NIST. FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), FIPS 205 (SLH-DSA). Approved August 13, 2024.
- NIST NCCoE. Migration to Post-Quantum Cryptography.
- S. Nakamoto. Bitcoin: A Peer-to-Peer Electronic Cash System. 2008.
- V. Buterin. Ethereum: A Next-Generation Smart Contract and DApp Platform. 2013.
- B. McGillion et al. Open-TEE – An Open Virtual Trusted Execution Environment. TrustCom, 2015. (PDF §3.1.)
- W. Dai et al. SBLWT: A Secure Blockchain Lightweight Wallet Based on TrustZone. IEEE Access, 2018. (PDF §3.2.)
- S. Houy, P. Schmid, A. Bartel. Security Aspects of Cryptocurrency Wallets – A Systematic Literature Review. ACM Computing Surveys, 2023. (PDF §3.3.)
- OP-TEE Documentation v23.12.3. Linaro/OP-TEE Project. (PDF §3.4.)
- M. Gentilal, P. Martins, L. Sousa. TrustZone-backed Bitcoin Wallet. CS2, 2017. (PDF §3.5.)
- ARM Ltd. ARM Security Technology: Building a Secure System using TrustZone Technology. White Paper, 2009.
- V. Costan, S. Devadas. Intel SGX Explained. IACR ePrint, 2016.
- AMD. SEV-SNP: Strengthening VM Isolation with Integrity Protection. White Paper, 2020.
- Android Open Source Project. A/B (Seamless) System Updates. 2025.
- Research constructor: “Quantum technologies: 75 questions for a research article” (Russian, 2026). Sections 1–12 were used, especially questions 1, 4–7, 10, 13–14, 17–18, 25–26, 31–32, 37–38, 43–44, 49–56, 63–64, 69–70.
