THWIXEL

×

THWIXEL: the technical description

For forty years the industry has answered its security problems with software: a patch, then a patch on the patch. We address the problem comprehensively in silicon, and in doing so open up two additional, separate ecosystems on a phone, tablet or laptop.

The display arrangement

A THWIXEL panel’s display driver IC accepts two independent video inputs: one from the host system, exactly as any device has today, and one from a separate Secure Processor that the host cannot read, write or reach.

The custom DDIC is pivotal to the strongest implementation, but it is not the whole of the invention, and much of what follows can be done without one: by a secure domain of the host, or by another party’s secure system, with assurance corresponding to the isolation achieved. Our filings cover it either way.

Three sources, one screen

Pixels can come from the Secure Processor, from the host, or from the Infinity Machine, and all three can be on the screen at once, anywhere within the display area. The one exception is the DSRB itself, the band at the perimeter: that is the Secure Processor’s alone, always, and nothing else may ever render there. What decides where each may render is not a fixed split of the panel but gating by the Secure Processor: it permits, positions and withholds. It is the arbiter for every pixel on the glass, inside the secure region and outside it. The host is a source, not a controller.

The DSRB, and why it cannot be faked

The Display Secure Region Border is how you see which pixels are which. A band at the perimeter of the display is permanently reserved to the Secure Processor and at no time available to any other source. Its ownership never changes; what changes is its appearance, and that appearance signals the state of everything within.

A secure window or icon carries a marker of its own, a border or a similar mark, and the user decides whether it is genuine by matching the colour and pattern of that marker against the DSRB. Anyone can draw a border: the host may put a marker around anything it likes. What the host cannot do is make its marker match the DSRB, because it cannot write in the band, cannot read what the band is showing, and cannot predict it. Only the Secure Processor can produce a marker that matches. The marker travels with its window as you move or resize it, and goes on matching throughout; one that does not match is not what it claims to be, and that is apparent on sight.

What matters is that the appearance changes: in colour, in pattern, in timing, or in any combination. How that change is arrived at is a matter of implementation, and it can as well be the user’s own choosing. The host can paint whatever it likes in its own pixels; what it cannot do is know what the band is showing at this instant, having no way to write there and no way to read it. The permanence of the reservation is what gives the comparison its force: were any other source ever able to paint in that band, it could reproduce the appearance, and every comparison against it would signify nothing.

Two kinds of indication are distinguished. One establishes that a region is genuine, and must be unpredictable. The other establishes whose it is, a bank’s channel or a government’s, and is fixed, because its job is to be recognised. They nest: a varying outer border, a fixed inner one naming the channel. A compromised banking provider cannot raise the government indication, and the reverse.

The indication need not be visual. It may be a light on the chassis wired only to the Secure Processor, rows of the panel beyond the area the host addresses, a tone, or a varying haptic pattern emitted as a finger crosses a region boundary, so a region can be verified by touch alone, with the display dark, covered, or turned away. For a user under observation, that is a capability no visible indication can offer.

The perimeter is also a control surface

Input at the perimeter is taken by the Secure Processor and is not delivered to the host, which neither observes that it happened nor can respond to it. Touch the band and a menu extends from it, bordered to match and varying with it, so you can see the menu is the Secure Processor’s and not something the host has drawn. It is reachable from every state of the display, including one in which the host holds the whole of the inner area, which the host can neither prevent nor imitate.

Because the band receives input the host cannot see, a confirmation of consequence can be taken there alone: the terms shown in a controlled region, the control to accept them in the perimeter, and neither reachable by the host. It also gives the Secure Processor a channel to the user that the host cannot suppress, alter, imitate or observe: for warnings, for a running view of what an automated agent is doing and the means to stop it, and for the coercion case noted below.

The Secure Processor

A distinct part with its own memory and its own path to the panel. It holds keys, performs attestation, and renders its own pixels. It is not a secure enclave inside the host SoC: an enclave still hands its output to the host to display. THWIXEL does not.

The Infinity Machine

The third source: an RP2350 or family part embedded in the device, MicroPython-programmable, independent of the host operating system, sharing the device’s resources including the display, and rendering under the same gating as everything else.

What follows from it

Anything that matters, whether a payment, a signature, an age check or an instruction to an AI agent, is rendered and confirmed in a secure region, and the touches and keystrokes meant for it reach the Secure Processor first, so what is private never enters the host. Because the Secure Processor can attest that specific pixels reached the glass on real hardware, delivery becomes verifiable: that is what THWANGO uses against ad fraud, and what lets content reach the glass without passing through anything able to capture it.

Control over AI agents

An agent acting for you can read, decide and act at machine speed, and the operating system it runs on has no way to tell a good instruction from a hijacked one. THWIXEL puts a gate in front of the acts that matter. An agent may propose a payment, a signature, a message or a change of settings; the proposal is rendered by the Secure Processor, in a secure region, and you confirm it there, on pixels the agent, its model and anything that has compromised the host cannot draw on or read. Thresholds and categories are yours to set. A compromised agent can spend nothing and send nothing you did not see and approve, and an agent that has gone wrong on its own is stopped by the same gate.

Monitoring what goes in and out

Touch and keypresses reach the Secure Processor first. What is destined for a secure region stays there and never enters the host at all, so there is no keystroke for host software to log. In the other direction, the Secure Processor watches the device’s sensors and its network: if the camera or the microphone comes on, it can say so on pixels nothing else can touch, so the indicator cannot be suppressed or faked the way a software light can. It is an independent witness inside the device, rather than the operating system reporting on itself.

Under coercion

An attacker standing beside the user, compelling them to log in, defeats every protection that assumes the attacker is remote. THWIXEL addresses that case directly, and it is a substantial part of the filed work. We do not set out the mechanism here.

Secure electronic voting

The ballot is completed inside a secure region, so no host process and no observer sees or alters it. It then travels in two cryptographic envelopes: an outer carrying identity, for eligibility and to prevent a second vote, and an inner carrying the vote, unreadable by whoever handles the outer. One authority knows who voted but not how; another knows how but not who; the same ballot runs simultaneously through independent chains operated by separate bodies, and the counts must agree. Excess reveals stuffing, a deficit reveals suppression, divergence reveals manipulation. The voter checks their own vote in the published record by a secret code, which shows the vote and not the voter.

Vote-buying is made worthless by the coercion mechanism above: the true vote is cast and counted under a code withheld until the count is final, while the buyer, demanding proof, is shown a result matching what they demanded and cannot tell the difference. The same architecture serves corporate, union and community ballots. It is covered in detail in one of the provisionals.

Published PCT application WO 2026/015933 (THWITRS), and a family of provisional applications covering the display arbitration, agent gatekeeping, age verification, coercion and the advertising architecture. The full IP list ›
The custom DDIC is the next major milestone.

×

What’s inside THWIXEL