[Exploration] Port SeedSigner to the NumWorks open graphing calculator?

[keegreil]

Hi! This idea is not SeedSigner specific, but since SS's ethos is multiplying easy access to Bitcoin signing devices, especially through repurposing other hardware, this seemed like the best community to put this idea to.

The NumWorks graphing calculator is an open source effort to enable high school math education. It's ARM7 based (STM32F730V8T6, same chip family as ColdCard and many other wallets).

Here's some of the advantages to using this as a signing device:

• Totally innocuous design and primary purpose. It's a calculator dude. Easy to hide in plain site, or take with anywhere.
• Some countries charge much lower import taxes for calculators than for other electronics devices, so this might be easier to ship around.
• There's a whole grassroots effort to get these devices widely distributed to high school kids that has nothing necessarily to do with Bitcoin per-se (though it'd be great to get kids into Bitcoin and math at the same time!)
• Much more limited, fully open source code base compared to SeedSigner and more complex devices. Smaller attack surface.
• OS already implements "Test Mode" (as in, for taking standardized tests like SAT without cheating), which deletes all recent inputs and memory. Also has a hard "Reset" which completely wipes the device and restores it to factory firmware. So possibly more verifiably "stateless".
• The ROM chip includes a "deep power down" mode, which prevents ANY data from being written to the chip at a hardware level, further assuring its "statelessness."
• Integrated Li-Po battery
• Huge beautiful screen for displaying QR codes.
• Tons of buttons for user input.
• The STM32 does have some security features which isolate sections of code and memory. I'm sure its nowhere near as good as a proper secure element.

Here's some disadvantages:

• Input would be hard. Either the PSBT would have to be sent over USB cable, or typed in manually. It might not be hard to type given the comfortable amount of buttons, but could take a while and be error prone.
• The private key would have to be typed by hand each time (no QR scanner input).
• OS code and "apps" are written in C++, so we couldn't port SS code directly. Perhaps we could port over Bitcoin Core wallet code however, or maybe even some of ColdCard's.
• Would require frequent USB connections, to re-load the Bitcoin wallet code, since the workflow would likely require a hard reset to factory firmware each time it is used. Possible mitigations are to fork the OS code itself and install custom firmware that includes the Bitcoin wallet app (so it doesn't dissappear each time, just the wallet data does).
• Careful checking of the signed transaction for correct addresses would be required before transmitting to the blockchain, since the stateless model and frequent USB connections means trust in the device itself should be very low. We're really just using it as a calculator to do the ECDSA and SHA-256 math and then nuking it each time, rather than do them by hand with pencil/paper. It's highly likely that it would be actively trying to leak the private key from the moment it is entered until we hit the "RESET" button.
• The display may be more susceptible to TEMPEST style attacks than SS (way out of my depth on this one, but I think ColdCard uses a very specific serial communication protocal for the Q1 to at least mitigate this a little bit).
• No secure element (but you already knew that. If SE is a must buy a CC).
• Compared to ColdCard Mark 4, this STM supports only AES encryption/decryption natively using its own hardware accelerator. The CC Mk4 "MCU implements SHA-256 and AES natively for quicker encryption/decryption."

So, that's my idea. Use the NumWorks graphing calculator as a covert signing device. I'd love to your thoughts, positive or negative! What major roadblocks am I missing? Would this be useful to the community?

[SeedSigner]

I like the creativity of this idea and I can tell you've put a good amount of thought into it. The display for numworks unfortunately doesn't look like it would offer an advantage over our current display; according to their specs the resolution is 320x240 pixels (https://www.numworks.com/specs/).

[keegreil]

Good catch on the resolution, I hadn't noticed that detail. I may tinker on it later this year if I get the chance, just to explore the possibility.