PostHeaderIcon Decode Locked NRF52833-QDAA-R7 Source Code: QFN40 512 KB Read

The NRF52833-QDAA-R7 carries the same nRF52833 silicon as the rest of the family, an Arm Cortex-M4 with FPU at 64 MHz, 512 KB of flash and 128 KB of RAM in a QFN40 5 x 5 mm body, and the R7 suffix only names the 7-inch reel container it ships on. When the flash has been pulled and the image is sitting there with no symbol table attached, the job is to decode locked NRF52833-QDAA-R7 source code into something that can be compiled again, and the difficulty is entirely in what the image does not say about itself.

The container code changes nothing

Nordic’s ordering codes put the shipping format in the last letters, so R is 13-inch tape and reel, R7 is a 7-inch reel and T is a tray, and the silicon, package, memory and temperature rating are identical across all three. A recovery on the NRF52833-QDAA-R7 therefore produces a binary that also fits the R and T parts.

What a 512 KB image gives you

A Cortex-M4 image starts with a vector table, and that table anchors everything after it. The initial stack pointer bounds the RAM the firmware was linked for, the reset vector points at the startup code, and each entry after that maps one exception or interrupt to its handler, which is how the interrupt structure becomes legible. From the handlers the call graph extends outward, and the branch targets inside each function mark its boundaries. Where the compiler has inlined a small helper, the boundaries blur, so the call graph is read as a graph. Literal pools holding string and constant data sit between functions, and each one marks the tail of the code block before it, fixing the function edges from the other direction.

Reading the reconstruction signals

The parts of the image that carry structure are not obvious at the start, so each is read against what the hardware can prove. Reading them together is what makes it possible to decode locked NRF52833-QDAA-R7 source code with reasonable confidence, because a constant pool carrying the PWM channel masks confirms a register address that was ambiguous on its own.

Image region What it yields Confidence
Vector table Interrupt structure, RAM bounds, reset path High
Call graph and branch targets Function boundaries and call relationships High
Literal pools Constants, strings, peripheral register masks Medium
Peripheral register addresses Which hardware each routine drives Medium
SoftDevice area Encrypted at rest, not decompilable None

Naming the peripherals from the addresses

The nRF52 peripheral set is wide and its register map is fixed, so a word read from a constant address identifies the block it came from. PWM, SAADC, SPIM, TWIM, UART, TIMER, RTC and GPIO accesses are all recognisable in the instruction stream by their address ranges, and that alone turns a column of constants into a map of what the firmware drives. The four 4-channel PWM units with EasyDMA, the 12-bit 200 ksps ADC, the 64-level comparator, the temperature sensor and the two UARTs each show up as their own cluster of accesses. Our dual probe and flying probe notes cover reading those pads.

Why this peripheral set complicates things

Three things make an nRF52 image harder to read back as source than a plain microcontroller. First there is the programmable peripheral interconnect, so a task on one peripheral can be triggered by an event on another without the CPU ever issuing a transfer, and that relationship lives in registers rather than in code, so it has to be reconstructed from access patterns. Second, EasyDMA means the descriptors are data rather than code, laid out in RAM and pointed at by registers, so a DMA-driven buffer never appears as instructions and has to be recovered as a structure. Third, the RF register block is largely invisible from the disassembly, because most radio configuration happens in hardware.

The region that cannot be rebuilt

One part of the 512 KB image is out of reach for a different reason. The SoftDevice Bluetooth stack area is delivered encrypted and is decrypted into RAM at boot, so a debugger read of that flash region returns ciphertext and it cannot be decompiled. It is not part of the reconstructed source, and no amount of work on the remainder will change that. We state it plainly in the report rather than shipping placeholders that pretend otherwise.

What you get back

You receive the flash binary image of the NRF52833-QDAA-R7, a disassembly-based source reconstruction organised into a compilable tree with the vector table, the interrupt map and the peripheral assignments documented, and a verification report recording which regions are proven code and which are unreachable encrypted content. The same bench handles the QDAA-R encrypted flash attack and the nRF52832 source rebuild, while a flash recovery pass confirms the image is whole first. The same technique covers the STM32G473RCT6 data on other silicon.

Comments are closed.

Categories