Attack Surface of Hardware Wallets

This article has three layers:

1. Layer 1: The Intuition: A safe can lock out a thief, but it cannot lock out a con artist.
2. Layer 2: Real-World Use: Kraken Security Labs' voltage glitch research, fake devices, phishing after data leaks, and the Ledger Recover controversy.
3. Layer 3: Security chips, fault injection, side-channels, firmware verification, and anti-tampering.

Layer 1: The Intuition

1.1 A safe can lock out a thief, but it cannot lock out a con artist

The last three are actually the most common.

Layer 2: Real-World Use

• Voltage Glitch Attack (Kraken Security Labs, 2020): Researchers demonstrated that on a specific hardware wallet without a secure element, given approximately 15 minutes of physical access, they could use precise voltage interference during chip startup to skip protection checks, read the encrypted seed, and then brute-force the PIN. The manufacturer's response: recommend users set a BIP-39 passphrase (which is not stored on the device) and use secure elements in new models.
• Fake and Tampered Devices: There have been cases where users purchased hardware wallets on e-commerce platforms that were "pre-configured with a recovery phrase," with a printed "recovery card" included in the packaging. Once the user followed the instructions and transferred funds, they were immediately stolen.
• Phishing after Data Leaks (2020): A hardware wallet manufacturer's customer database (names, addresses, phone numbers) was leaked. Subsequently, users received a large number of emails, SMS messages, and even fake devices sent to their homes impersonating the official brand to induce users to enter their recovery phrase.
• Ledger Recover Controversy (2023): The manufacturer introduced an optional cloud backup service for recovery phrases, sparking community debate: is the promise that "the private key never leaves the device" a hardware guarantee or a firmware policy? This made many people realize for the first time: firmware can be upgraded, so the security boundary is defined by the firmware.

Layer 3: Principles and Engineering

3.1 Secure Element

This creates a trade-off: Open-source auditability vs. physical attack resistance. Different manufacturers make different choices; some use two chips simultaneously.

3.2 Fault Injection

Creating interference at critical moments of chip operation to make it "calculate one step incorrectly":

• Voltage Glitch: Momentarily dropping the supply voltage;
• Clock Glitch: Inserting an excessively short clock cycle;
• Electromagnetic Pulse, Laser: Precisely hitting a specific area of the chip.

The goal is usually to skip a comparison instruction (e.g., "Is the PIN correct?" or "Is read protection enabled?"). Defense: redundant checks, random delays, self-destruction of critical data upon anomaly detection, and writing critical decisions as "pass only if multiple calculations match."

3.3 Side-channel

Instead of reading data directly, information is inferred through power consumption, electromagnetic radiation, or execution time while secrets are being processed. For example, if an elliptic curve scalar multiplication performs an "add" or "no add" operation based on each bit of the private key, the private key's bits can be read directly from the power consumption curve. The defense is constant-time, constant-power implementation (the sequence of operations remains the same regardless of the key) and the use of random masking.

3.4 Firmware and Supply Chain

3.5 The screen is the last line of defense

The most important security feature of a hardware wallet is actually that independent small screen: it is a display channel that is not controlled by the computer (see "Out-of-Band Verification"). However, if the screen only displays a string of hashes or hexadecimal data, the user can only perform blind signing, and this line of defense fails (see "Blind Signing and What You See Is What You Sign," the Bybit incident, and frontend integrity). Therefore, choose a device that supports clear signing (decoding transactions into human-readable language) and actually read the content on the screen.

Summary