Invisible bits: hiding secret messages in SRAM's analog domain

Encryption hides what a message says but shouts that a secret exists, which is exactly what gets a traveler coerced into handing over the key at a border; this work from Virginia Tech hides secret data not in a file but in the analog physics of ordinary SRAM chips — subtly biasing how strongly individual memory cells prefer a 0 or a 1 — so the message is invisible to any normal inspection of the device's contents and recoverable only by someone who knows to look at the memory's power-up behavior, giving the owner plausible deniability that any secret is there at all.

The argument in one glance

Problem

Encryption hides a message but advertises that it exists

Electronic devices are increasingly inspected by authorities, and encryption is the wrong tool against a coercive inspector: it hides what a message says, but the encrypted data is conspicuous, so it announces that a secret exists. An adversary who sees encrypted data can simply coerce the owner into surrendering the key. What is needed is not stronger encryption but a way to hide that any secret is present at all.
What people did

Use steganography to hide the message in plain sight

Steganography hides a secret message inside something that looks ordinary, so the message's very existence is concealed and the owner gains plausible deniability. Classic techniques tuck data into the unnoticed corners of images, audio, or files — places a casual look would pass over.
Why it fell short

Ordinary hiding places are on the digital surface, where inspectors look

The trouble is that file-based steganography still lives in the device's digital contents — the very thing an inspection dumps and scrutinizes. A carrier file that is examined closely can betray its payload, and any hiding place that shows up in a normal read of storage is exposed to anyone who thinks to look. A truly deniable hiding place would need to be invisible to a digital inspection of the device and yet not interfere with the device's normal operation.
This paper's idea

Hide the message in the analog domain of ordinary SRAM

Invisible Bits hides secret data beneath the digital surface, in the analog physics of the SRAM chips a device already contains. Every SRAM cell has a subtle analog preference for a 0 or a 1, arising from manufacturing variation; Invisible Bits encodes the message by deliberately biasing chosen cells' preferences. The SRAM stays fully usable for normal storage, the bias is indistinguishable from ordinary manufacturing variation under a digital read, and the message is recoverable only by examining the cells' analog behavior — such as the memory's power-up values.

Why this problem is worth solving

To see why this hiding place is so effective, it helps to understand a hidden layer inside ordinary memory.

SRAM (static random-access memory) is the fast on-chip memory found in nearly every processor and device. Each SRAM cell is a tiny bistable circuit — a pair of cross-coupled inverters — that holds either a 0 or a 1. That digital value is the whole story as far as normal use is concerned. But underneath it lies an analog reality: no two transistors are manufactured identically, so the two halves of each cell are slightly mismatched, and that mismatch gives every cell an intrinsic preference. Some cells lean strongly toward 0, some strongly toward 1, and some are nearly balanced. This shows up most clearly when an SRAM is first powered on before anything is written to it: each cell settles into its preferred value, producing a pattern that is unique to that physical chip — the basis of SRAM physical unclonable functions (PUFs) used for device fingerprinting. Crucially, the strength of a cell’s preference is a continuous, analog quantity that is completely invisible to a normal digital read, which sees only the 0 or 1 currently stored. This analog domain — how strongly each cell leans, not merely which way — is a hidden dimension beneath the digital abstraction, and it is where Invisible Bits stores its secret.

The reason to hide there rather than encrypt comes down to a specific and nasty threat.

Encryption and steganography solve different problems. Encryption makes a message unreadable without a key, but it does nothing to conceal that a message exists — a blob of high-entropy ciphertext on a device is conspicuous, practically waving a flag that says “a secret is here.” Against a passive eavesdropper that is fine, but against an adversary with physical power over the owner it is a liability: seeing the encrypted data, the adversary knows exactly what to demand, and can coerce the owner — legally or otherwise — into surrendering the key. Border inspections and device seizures are precisely this scenario. Steganography attacks the other half of the problem: it hides the very existence of the message, embedding it in a carrier that looks completely ordinary, so the owner can plausibly deny that any secret is present at all. If the inspector finds nothing that even looks like a secret, there is nothing to coerce. The challenge is that the carrier must be genuinely innocuous — indistinguishable from the normal state of the device — and must survive the device continuing to be used, or the ruse falls apart. Finding such a carrier is exactly what makes SRAM’s analog domain valuable.

What people did before, and where it broke

Encryption. Make the message unreadable without a key. Where it broke. It hides content but not existence; a conspicuous ciphertext tells an adversary a secret is present and lets them coerce the owner for the key.
File-based steganography. Hide the message in the unnoticed parts of an image, audio file, or document. Where it broke. The carrier still lives in the device’s digital contents, which an inspection dumps and scrutinizes, so a closely examined carrier can betray its payload and the hiding place is on the surface where inspectors look.
Hiding in unused storage or slack space. Tuck data into filesystem slack, hidden partitions, or unallocated blocks. Where it broke. These are still digital artifacts that forensic inspection can find, and their presence is anomalous once someone knows to look, offering weak deniability.
The gap. No hiding place was simultaneously invisible to a digital inspection of the device, undisturbed by the device’s normal use, and available on commodity hardware without adding components. A carrier beneath the digital abstraction, in the analog physics of memory the device already has, is what Invisible Bits provides.

How Invisible Bits works

From a conspicuous ciphertext to a message hidden in the memory's physics

↓ →
new
↓ →
new
↓ →
new
↓ →
new
↓ →
new

The idea

Reach beneath the digital abstraction into the analog physics of the same chip. A digital read of SRAM sees only 0s and 1s, but the same silicon carries a continuous analog reality — how strongly each cell leans — and Invisible Bits computes and stores in that hidden layer the digital view discards. Descending below the clean digital abstraction to exploit the underlying analog behavior of ordinary CMOS, which the digital layer renders invisible, is exactly the move the temporal-arithmetic work covered elsewhere on this site makes, computing in the analog timing domain of the very same logic gates that the digital abstraction treats as instantaneous.

Store the message in the physical medium’s own properties, not in logical contents. The secret is not a pattern of stored bits but a pattern of physical biases in the memory cells themselves, encoded into the substrate rather than written on top of it. Encoding information directly into the physical characteristics of a storage medium, working with its native physics rather than a logical layer above it, is the same approach the DNA-storage encoding work covered elsewhere on this site takes, writing data into the physical molecules of a medium under the constraints that physics imposes.

Root the security in a low-level hardware property rather than a software policy. The deniability comes not from a program or an access rule but from a physical fact — that a deliberately biased cell is indistinguishable from a naturally biased one — which no software inspection can see through. Grounding a security guarantee in the hardware’s own behavior rather than trusting a software layer is the same discipline the capability-hardware memory-safety work covered elsewhere on this site follows, enforcing its property in the low-level machinery of the processor where software cannot quietly bypass it.

Following a secret through Invisible Bits

Hiding and recovering a message in SRAM

Tracing one secret shows why the hiding place survives inspection.

Step 1 of 5

Choose ordinary SRAM as the hiding place

The owner needs to carry a secret through an inspection without leaving any evidence that a secret exists. Rather than encrypt a file, they use SRAM the device already contains as the carrier.

No extra hardware is added; the hiding place is commodity memory.

Why it works

There is nothing to coerce. Encryption leaves an obvious blob that proves a secret exists → so an adversary can coerce the owner for the key → but hiding in the analog domain leaves nothing to point at, so there is nothing to demand.
The message hides in natural variation. Every SRAM cell has an analog preference from manufacturing variation → so a deliberately biased cell is indistinguishable from a naturally biased one → so the hidden message looks like the ordinary device-to-device variation of any chip.
The carrier keeps working. The analog bias sits beneath the digital value → so normal reads and writes are unaffected and the memory stays fully usable → so the hiding place does not interfere with the device’s real work.
Inspection looks in the wrong place. A normal inspection dumps the digital contents → but the message is not in the digital contents, only in the analog domain → so a routine read of the memory reveals nothing.
Only the initiate can read it. Recovery needs a specific, non-standard measurement of the cells’ analog state → so only someone who knows the scheme can retrieve the message → so the data stays secret even on a fully inspected device.

What it costs (trade-offs and scope)

Writing the analog domain is slow and wears the hardware. Biasing a cell’s preference means persistently stressing it, which takes time and gradually ages the memory, so the hiding channel is low-bandwidth and not free of physical cost.
Analog storage is noisy, so recovery needs care. The analog preferences are subtle and affected by temperature, voltage, and noise, so reading the hidden bits reliably requires error tolerance, which limits how much can be stored and how dependably.
The secrecy assumes the adversary does not know the scheme. Plausible deniability holds as long as inspectors do not think to measure the analog domain; an adversary who specifically knows to look for this technique could probe for it, so the hiding place is not unconditionally secure.
It requires physical access to write and to read. Both biasing the cells and recovering the message need a specific procedure on the device, so this is a scheme for hardware the owner controls, not a remote communication channel.
The capacity is small. Hiding data in analog cell preferences yields far less room than ordinary storage, so it suits short secrets — keys, credentials, brief messages — rather than bulk data.
It is tied to the physics of SRAM. The technique exploits the specific behavior of SRAM cells; other memory technologies would need their own analysis, and the exact biasing mechanism, capacity, and robustness are detailed in the full paper.

The evaluation

Three ways to protect a secret message on a device

Property	Encryption	File steganography	Invisible Bits
Hides the message content	Yes	Yes	Yes
Hides that a secret exists		Partly
Where it hides	In a visible ciphertext
Deniability under inspection	None	Weak if the file is scrutinized

Three ways to protect a secret on an inspected device. Encryption hides content but advertises that a secret exists, inviting coercion; file steganography hides on the digital surface an inspection scrutinizes; Invisible Bits hides beneath the digital abstraction, in the analog preferences of on-chip SRAM cells, where the message is invisible to a normal read, indistinguishable from ordinary manufacturing variation, and does not disturb the memory's normal use — giving strong plausible deniability. (This explainer is drawn from the paper's abstract and standard background on SRAM and steganography; the full text, including the exact biasing mechanism, capacity, and robustness measurements, was not available.)Underlined values explain themselves when clicked.

Where this stops working

What to keep in mind

Invisible Bits is a striking idea in hardware security: it answers a threat encryption cannot — a coercive inspector who will extract a key once they know a secret exists — by hiding the message where no digital inspection looks, in the analog physics of ordinary SRAM. Because every cell already has a manufacturing-induced preference, a deliberately biased cell blends into that natural variation, the memory keeps working normally, and only someone who knows to measure the analog domain (for example the power-up pattern) can recover the message, giving genuine plausible deniability. Read it with its scope. Writing to the analog domain is slow and ages the hardware, so the channel is low-bandwidth and suits short secrets, not bulk data; the analog state is noisy, so recovery needs error tolerance; the deniability rests on the adversary not knowing to probe for this specific technique; and it needs physical access to write and read. Importantly, this page is written from the paper’s brief abstract and standard background on SRAM physical unclonable functions and steganography, because the full text was behind an inaccessible paywall; the description of the mechanism (biasing cells’ analog preferences and recovering via their analog behavior) is a faithful reading of the title and SRAM physics, but the exact biasing procedure, the storage capacity, the error rates, and any defenses are detailed in the paper and are not reproduced or invented here. The durable lesson is that the clean digital abstraction hides a rich analog reality beneath it, and that reality can be a place to store, to compute, or — as here — to hide.

Background

Why hardware's physical layer is a security frontier

It helps to see Invisible Bits as one move in a broader game being played at the physical layer of hardware, beneath the tidy world of bits. That layer is full of exploitable physics: the same manufacturing variation that fingerprints a chip as a physical unclonable function is here turned into a place to hide data; elsewhere, subtle physical effects leak secrets through side channels, or let attackers flip bits they should not be able to touch. Both attackers and defenders increasingly operate there, because it is invisible to software and to the digital abstractions most security tools reason about. The work on hardware-assisted memory safety covered elsewhere on this site sits on the defensive side of the same frontier, pushing security guarantees down into the processor’s low-level machinery where software cannot quietly circumvent them. The broad lesson is that the physical substrate of computing is not a neutral, faithful implementation of the digital model but a rich medium with its own behavior — and that some of the most powerful capabilities and most dangerous threats live precisely in the gap between what the hardware physically does and what the digital abstraction admits it does.

If you remember only three things

1. Encryption hides what a message says but not that it exists — a conspicuous ciphertext lets a coercive adversary (a border inspector, a device seizure) demand the key. Steganography instead hides the existence of the message, giving plausible deniability, but classic file-based steganography still lives in the device's digital contents, which an inspection scrutinizes. Invisible Bits provides a hiding place beneath the digital abstraction: the analog domain of ordinary SRAM, where each cell has an intrinsic, manufacturing-induced preference for a 0 or a 1 that is invisible to a normal digital read.
2. Invisible Bits encodes a secret by deliberately biasing chosen SRAM cells' analog preferences, so the message lives in how strongly cells lean rather than in the bits they store. The SRAM remains fully usable for normal storage (ordinary reads and writes are unaffected), a digital dump reveals nothing because a biased cell is indistinguishable from a naturally biased one, and the message is recovered only by a non-standard measurement of the cells' analog behavior — such as reading the SRAM's power-up pattern. The result is strong plausible deniability: an inspector who dumps the memory sees only ordinary manufacturing variation, with nothing that even looks like a secret to coerce a key from.
3. Its scope: writing the analog domain is slow and gradually ages the hardware, so the channel is low-bandwidth and suits short secrets (keys, credentials, brief messages) rather than bulk data; the analog state is noisy, so recovery needs error tolerance; the deniability holds only against an adversary who does not know to probe for this specific technique; and it requires physical access to write and read, on hardware the owner controls. This explainer is written from the paper's brief abstract and standard SRAM-PUF and steganography background, because the full text was unavailable — the exact biasing mechanism, capacity, error rates, and defenses are in the paper and are not reproduced here.

Check yourself

Check your understanding

1. Why is encryption the wrong tool against an adversary who can inspect and coerce?
2. What is the "analog domain" of SRAM that Invisible Bits hides data in?
3. How does Invisible Bits hide and recover a message, and why is it deniable?
4. Which statement best reflects the scope and limitations of this work?

Sources

The paper is “Invisible bits: hiding secret messages in SRAM’s analog domain” by Jubayer Mahmod and Matthew Hicks, published at ASPLOS 2022, DOI 10.1145/3503222.3507756. The authors are at Virginia Tech.

This explainer is marked as a draft because the full text was not accessible: the paper is gold open access (Creative Commons) on the ACM Digital Library, but that page was protected by an anti-bot challenge that blocked automated retrieval, and no arXiv preprint, author-hosted copy, or open repository version could be found; only a brief abstract was available. The account above is written from that abstract together with standard, well-established background on hardware security — SRAM cell structure and power-up behavior, the manufacturing-variation-induced preferences that underlie SRAM physical unclonable functions, and the distinction between encryption (which hides content but not existence, inviting coercion) and steganography (which hides existence for plausible deniability). The specific claims from the abstract are that electronic devices are increasingly inspected, that encryption calls attention to secrets and invites coercion, and that steganographic techniques hide secret messages in plain sight to provide plausible deniability and remove the threat of coercion. The description of Invisible Bits’s mechanism — encoding a message by biasing SRAM cells’ analog preferences, leaving the memory usable and the message invisible to a digital read, and recovering it via the cells’ analog behavior such as the power-up pattern — is a faithful reading of the paper’s title (“hiding secret messages in SRAM’s analog domain”) combined with the established physics of SRAM cells; however, the exact biasing procedure, the storage capacity, the error rates and robustness, and any proposed defenses are detailed in the full paper and are not reproduced or invented here, and should be confirmed against the paper itself.