Ryan Walker
Because we all know the best way to keep your data safe is by blowing it up, right?
Part 1
Well, for most of us, the real answer lies with encryption. But consider this: there are certain countries where it is dangerous to be a journalist. Now pretend you’re investigating political corruption in one of these countries; in such cases, it may not be safe for you to be found with an encrypted drive.
My idea is to build a USB drive that is cosmetically and functionally identical to your run-of-the-mill USB. But there is one difference: if you plug it in normally, it doesn’t produce any data.
Now, what’s something a sane person would never do before plugging in an ordinary flash drive? Lick their fingers.
That’s right; The drive will have hidden electrodes that measure the resistance of the finger plugging it in. A finger is around 1.5MΩ, but wet fingers are around 500kΩ. When the device boots, the drive will appear blank if the resistance between the electrode pair is higher than a threshold. It’s not the prettiest system, but I think it strikes a nice balance of ridiculousness and functionality. The germaphobes among us can run their fingers under the tap.
The Design
A flash drive is a relatively simple design electrically. For the first version, I’m going to target USB2.0 speeds.
The typical flash drive is composed of a USB controller (blue) connected to a NAND flash chip (red). The flash chip holds all the data, while the controller contains a USB front end and logic to interface with the flash chip. To achieve my desired functionality, ill use a USB controller with a small microcontroller to read the electrodes and inhibit the flash chip if necessary.
Understanding component economics is essential when designing hardware. The flash chip is generic and can be used in anything: smart TVs, computer BIOS, cars, you name it. However, the USB controller is an **application-specific **component for flash drives.
Application-specific ICs come around when there’s a huge market, and you need to squeeze margins. There might only be a few dozen flash drive companies worldwide, and the engineering is stale. The lion’s share of the volume comes from several huge factories with razor-thin margins. There’s no hot new flash drive startup that’s going to disrupt the market. Digikey or Mouser isn’t the place to look for sourcing application-specific components.
I scoured the internet for flash drive teardowns, searched the text on the chips, and found a gold mine: a flash drive database that lists part numbers for several USB controller ICs. I then checked to see if I could get a datasheet and a vendor — I settled on the SM3257EN.
SM3257EN Block diagram
This chip looks like it should do the trick; the datasheet is well written and has enough information. I created the part in Kicad and imported it with a NAND flash chip and USB jack.
I now have to make the device hide the data unless the user licks their fingers.
A chip enable (CE) signal from the USB controller is designed to connect straight to the flash chip. When this signal is low, the flash chip will enable. I will use an "or gate" with my control signal to turn force the memory off.
When my inhibit signal is high, it doesn’t matter if the status of CE. The output of the gate will be high, which disables the flash memory.
As for controlling the inhibit signal itself, I’m going to use an ATTINY24, and a transconductance amplifier hooked up to the electrodes. I’ll get more into that next post.
The Mission
I’m trying to build hardware that solves problems and builds a community. If you think you might have a use for the hardware in this post or would like to help out, I would love to hear from you. I’ve created a Discord server with like-minded people! Everything is on github!
Part 2
I’m building an open-source USB drive with a hidden self-destruct feature. Say goodbye to your data if you don’t lick your fingers before plugging it in. Its target customers are journalists in anti-privacy countries and security researchers. Part One Here
This post outlines the design process and challenges. I’ll discuss my bench prototyping, the schematic, layout and mechanical sourcing.
A lot of the responses to the last post pointed out that the device doesn’t actually self-destruct. In response to these comments, the device will now have two modes: just hiding the data and a full self-destruct (FSD).
Enclosure
I found a shop overseas selling USB enclosures without the internals, which saves me from designing an enclosure from scratch. I’ve never designed an injection moulded enclosure before, and today won’t be the day.
USB Drive Enclosures
Out of the four samples I received, I settled on the black one (far left in the picture above). Unfortunately, the vendor could not give me 3D CAD files, just DXF. If you don’t know, DXF is a simple flat drawing; not ideal, but better than nothing.
Drawing of Enclosure
I took these DXF drawings, imported them into FreeCAD, and then extruded them into some 3D models. This was a strange way to draw a model; the DXF lines were converted to undimensioned FreeCAD sketch objects and then extruded. I imported the PCB and ended up here:
Device Render
Hopefully, the final build will come out looking something like that.
FSD: Full Self Destruct
My goal is to build a discrete device; if the cops snatch a journalist in a non-privacy country, they shouldn’t think twice about a loose USB drive. When they plug it in, the device shouldn’t explode, melt, release corrosive material or do anything else insane (even though that would probably make a more exciting blog post). It should quietly destroy itself beyond repair.
My solution for this is overloading the flash memory voltage rail. I’ll have to say, this is the first time I’ve ever actually looked at the absolute maximum ratings of a component.
The part needs to be pushed over 4.6V in order to be completely disabled. I can use a simple voltage doubler off the 5V line to do this.
The operation of this circuit is pretty simple. When Distruct_PWM is low, Ca will charge to 4.3V, which is 5V minus the 0.7V drop over the diode. When I set Distruct_PWM high from the MCU, this puts the bottom of Ca at 5, giving a total potential of 9.3V. This flows into Cb and gets trapped for the next cycle. When you want to dump the energy into the flash IC, enable Q1 and say goodbye to those cute dog pics.
Sensing Circuit
OK, so how will the device know if the user has licked their fingers or not, and hence whether or not to destroy the data? This can be done with a bioimpedance measurement, i.e., a measurement of the resistance of the user’s skin. If the skin’s resistance is low (500k or less), we can assume their fingers are wet. There’s probably a fancy chip to measure bioimpedance, but since the chip shortage, I’ve been inclined to use generic components.
I pinched this circuit from The Art of Electronics; it’s a current supply. Let’s discuss how it works. U1 is a voltage reference; as long as it gets enough current to operate, there will be 2.5V dropped over it. The cathode is connected to R2, while the anode is connected to the non-inverting input of the opamp. The output of the opamp will supply the current required to hold the two inputs at the same voltage. Therefore we can assume there is 2.5V dropped over R2 as well. The current through R2 is then
Our electrodes are connected across J1, and the voltage at Vx and Vs is again determined by Ohm’s law:
So we can use these two equations to solve for the load resistance, RL:
This is what we’ll use to calculate skin resistance.
Microcontroller
I chose a simple attiny25 for the brains of the project. It’s a step down from the 32bit ARM chips I typically use. It was refreshing to fit the entire application into 55 lines of code with only one required header.
I configured a PWM channel to test the code. I read and ADC pin and output the voltage as a duty cycle on the PWM pin. Using my multimeter on DC mode, I read back the voltage I was sampling.
#include <avr/io.h>
// PWM PA6
// ADC PA1
// LED PA2
const float R2 = 747e3; // R2 in schematic
const float Rth = 0.4e6; // If r > 1Mohm, finger not wet.
void init_pwm() {
DDRA |= (1 << PA6); // PA6 as output
OCR1A = 0x0000;
TCCR1A |= (1 << COM1A1); // set non-inverting mode
TCCR1A |= (1 << WGM11) | (1 << WGM10); // set 10bit phase corrected PWM Mode
TCCR1B |= (1 << CS11); // set prescaler to 8 and starts PWM
}
void set_pwm(float voltage) {
OCR1A = ( voltage / 5 )* 0x400;
}
void init_adc() {
ADMUX &= ~(1 << REFS0); // Use 5V as reference
ADMUX &= ~(1 << REFS1);
ADMUX |= (1 << MUX0); // Use ADC1 (PA1)
ADCSRB |= (1 << ADLAR); // Left adjusted (8bit) operation alright
ADCSRA |= (1 << ADEN); // Enable ADC
}
float read_adc() {
float voltage;
ADCSRA |= (1 << ADSC);
while(ADCSRA & (1 << ADSC)) {};
voltage = ADCH;
voltage = voltage * 5 / 0xff;
return voltage;
}
int main(void) {
DDRA = 1 << PA2; // LED
init_pwm();
init_adc();
while(1)
{
float v, r = 0;
v = read_adc();
set_pwm(v);
r = v * (R2/2.5);
(r > Rth) ? (PORTA &= ~(1 << PA2)) : (PORTA |= 1 << PA2);
}
}
The Build
When working for a client, I don’t breadboard. It’s too expensive and doesn’t make sense with modern SMD components. My usual workflow is to go from simulation straight to schematic and board design. But for this project, I reached for the breadboard. I did this in order to derisk the bioimpedance circuit I discussed above. It felt like I was back in school again, nostalgic about DIP components and paper schematics.
Let there be light!
The logic works fine. All of the learnings from this prototype were integrated into the final design.
The next and final post will be documenting the entire build, I’m hoping to then get a small crowdfunding campaign launched to build several of these devices for the community. Stay in touch and happy hacking!