Key Storage Is a Long-Term Challenge: Why CrossBar Builds on ReRAM

What years of storage do to the memory inside a hardware wallet, and why CrossBar chose a different kind.

Updated on
Key Storage Is a Long-Term Challenge: Why CrossBar Builds on ReRAM

A hardware wallet spends most of its life doing nothing. It is loaded once, put away, and expected to be correct when it is next needed, which may be years later. Whether the key is still intact at that point depends on the memory holding it, and that is a question buyers rarely ask.

Almost every hardware wallet on the market stores key material in Flash or EEPROM, and commercial Flash is typically rated for data retention on the order of ten years. CrossBar builds on a different technology, Resistive Random Access Memory (ReRAM), which the company has spent more than a decade developing. This post explains why.

1. Durable: Why Charge-Based Memory Has a Time Limit

Flash and EEPROM hold a bit as a small electric charge, trapped inside the cell. The data lasts only as long as the charge does.

Charge does not stay put. It leaks slowly at room temperature and faster as temperature rises, which is why a retention rating always comes with a temperature attached to it. Radiation drains it as well, whether from X-ray inspection of boards or luggage, from alpha particles given off by packaging materials, or from the higher exposure that comes with altitude.

Charge can also be measured, which means it can be read. Published attacks have recovered Flash contents using an electron microscope, read Flash through the light a chip gives off as it works, flipped individual bits with a laser, and worked out keys from the power a chip draws while reading memory. For a device that may sit in a drawer or a safe for a decade, none of this is theoretical.

2. Structural: Storing a Bit as Resistance Rather Than Charge

This memory holds a bit as resistance instead. Programming a cell moves a small number of metal ions inside the chip, forming or dissolving an extremely fine conductive thread. The cell is left either low resistance or high resistance, and that state is a physical structure rather than a pool of electrons.

The failure mechanism changes as a result. A charge-based cell degrades as its charge drains away. A thread has no charge to lose.

3. Projected: What the Retention Figures Show

All memory of this kind degrades faster when hot and slower when cool, but not every technology at the same rate. Published laboratory figures show ours slowing down more than Flash does as conditions cool.

Modeling two parts each qualified for ten years at 125 °C, the Flash part projects to roughly 70 years of retention at 100 °C, while ours projects to roughly 188. The gap continues to widen as temperature falls, and at room temperature the same model places the advantage at well over an order of magnitude.

These are model projections and do not constitute a warranty. What they establish is direction: the closer conditions get to those of ordinary storage, the larger the margin becomes.

Radiation follows the same pattern. With no charge to drain, our memory keeps working well past the levels at which some charge-based memory begins to lose data, beyond X-ray screening, air travel, or harsh industrial environments, and into the range normally associated with space-qualified hardware.

4. Tested: What an Independent Laboratory Found

Retention was the property CrossBar set out to improve. Resistance to physical attack followed from the same physics, because most published methods for extracting data from Flash begin by locating charge, and here there is no charge to locate. The thread also sits buried between the metal layers of the chip, shielded from imaging on either side, and even a high-resolution electron microscope cannot reliably tell which state it is in.

Reasoning of that kind is worth testing rather than asserting. CrossBar commissioned MicroNet Solutions, an independent laboratory, to attempt extraction of the contents of an array using any technique available, including removing its layers one by one, cutting into it with an ion beam, imaging it, analyzing its power, and destroying it entirely. They were unable to recover the data.

5. Layered: Where the Memory Sits in the Device

In CrossBar's key device, its stablecoin wallet, this memory holds the user's key material, the device's own identity, and the counters that enforce PIN and fingerprint attempt limits. Those counters move in one direction only and cannot be reset by cutting power.

Each unit carries an identity that the app verifies during pairing, so a tampered or counterfeit device is rejected before it ever holds a key. Firmware updates arrive signed, over the air from the user's phone, and cannot be rolled back to an earlier version. Every transaction is shown on the device's own screen and confirmed there.

Memory is not the whole of the device's security, and CrossBar would not present it that way. The protection is layered, and what the memory contributes is the foundation beneath those layers.

Closing: Built for the Years It Will Spend Unpowered

A key device has to remain correct through long periods in which nothing happens to it at all. Heat, radiation, and time act on a secret stored as charge in ways they do not act on one stored as structure. That is the property CrossBar chose to secure first.

Secure packaging, a supply chain with attestation at every step, and power design for a device that may sit unused for long stretches all matter as well, and CrossBar will address each of them in future posts.


—The CrossBar Team

Technical detail: ReRAM stores each bit as the resistance of a conductive filament roughly a nanometre across rather than as trapped charge. Retention figures are Arrhenius-model projections based on published activation energies, approximately 1.0 eV for NOR Flash (NXP engineering bulletin) and approximately 1.5 eV for ReRAM (IEEE IEDM), and do not constitute a product warranty. On radiation, some charge-based memories begin losing data at around 25 krad, while CrossBar ReRAM tolerates levels in the hundreds of krad. The attacks referenced are scanning electron microscopy voltage contrast, photon emission analysis, laser fault injection, and power side-channel analysis; ReRAM writes at around 3 V rather than 10 V or more and performs no block erase, which reduces the available power signature. Independent extraction testing was performed by MicroNet Solutions, Inc. under contract to CrossBar, using de-processing, focused ion beam milling, electron imaging, power analysis, and fully destructive methods. The device described is built around the Daric Secure Processing Unit, a CrossBar chip manufactured at TSMC.