Experience
Engineering experience
Our engineers have worked on hardware and embedded systems across a range of silicon and platforms. Most of that work was done under prior employment or as independent contractors, before Mycronex Labs existed as a company. Each entry below is attributed to where the work happened — never to Mycronex Labs as a company.
Entries
A real-time voice platform, Raspberry Pi to Rockchip
Attribution
Work by our embedded lead, carried out in a prior senior engineering role rather than for Mycronex Labs. Described here as team experience, not a company project.
Problem
A voice product built on Raspberry Pi needed to move to Rockchip RK3566 and RK3568 hardware without losing any peripheral support, and it had to wake up fast enough to feel immediate.
Constraint
Real-time audio leaves no headroom for latency, and a platform migration puts every driver, device tree and boot assumption back in question at once.
What we built
Customised Debian and Yocto images for both Raspberry Pi and the RK35xx SBCs. Kernel drivers and user-space APIs for the XMOS XVF3800 voice processor over I²C, SPI and I²S. Driver and device tree overlay integration for the MAX98396 amplifier. An ALSA SoC pipeline wired through PulseAudio and PipeWire for low-latency processing. Kernel and device tree work to bring full peripheral compatibility across to the new silicon.
Outcome
Boot time was cut substantially through kernel defconfig tuning, custom DTS and DTBO work, systemd service pruning and init sequence optimisation. The platform was validated at system level for real-time behaviour and audio fidelity.
A TCP/IP stack for constrained IoT devices
Attribution
Work by our embedded lead in a prior R&D role, not a Mycronex Labs engagement. Team experience.
Problem
A family of IoT devices needed to talk to each other and to the network across three very different targets — an STM32 microcontroller, a Compute Module 4, and full Linux.
Constraint
The microcontroller end has kilobytes of RAM, not megabytes, so a general-purpose stack was not an option and the same protocol behaviour still had to hold on all three.
What we built
A TCP/IP protocol stack designed for constrained embedded devices across STM32, CM4 and Linux. ZigBee (802.15.4), AODV and OLSR were customised and simulated for the radio side, with mesh and AP-client modes brought up on Raspberry Pi under Linux and OpenWRT driver modules integrated for mesh operation. NRC7292 Wi-Fi HaLow (802.11ah) and Silex NEWAH modules were taken on hands-on.
Outcome
Devices interoperated across all three targets on one protocol implementation, with DHCP, DNS and wpa_supplicant configured for mesh deployment.
A Jetson Orin Nano BSP with a production validation suite
Attribution
Work by our embedded lead under prior employment, not a Mycronex Labs engagement. Team experience.
Problem
An NVIDIA Jetson Orin Nano carrier needed a board support package under JetPack 6.2, a set of sensors and interfaces brought up on it, and a repeatable way to flash and verify units.
Constraint
Production flashing has to be both automated and secure, and every peripheral has to be proven on the real board rather than assumed from a datasheet.
What we built
Kernel and BSP customisation under JetPack 6.2, with GPIO pin multiplexing, device tree overlays and backported upstream drivers. Custom flashing scripts with an encrypted flashing workflow using LUKS. Peripherals validated on hardware: the ADV7812M CSI camera interface, a MAX17205 battery fuel gauge, INA power monitors, an accelerometer and a magnetometer.
Outcome
The board could be provisioned and verified repeatably as a production step rather than a bench exercise, with each interface confirmed on real hardware.
A multi-microphone audio board, schematic through test
Attribution
Work by our hardware lead, delivered independently before Mycronex Labs was formed. Team experience.
Problem
A voice product needed a compact board that could capture usable audio from several microphones at once and drive a speaker, without the analog front end picking up noise from everything around it.
Constraint
Microphone inputs are among the most noise-sensitive circuits on a board, and they had to share it with switching power and digital logic in a small outline.
What we built
A four-layer multi-microphone audio board taken from schematic capture through layout, with the layer stack used to keep the sensitive analog sections isolated. The bare board was milled in-house on the lab's CNC, then assembled and brought up on the bench. Related work covered the full path for an AI assistant smart speaker: schematic, PCB layout, 3D enclosure design, assembly, prototype build and test.
Outcome
The board was assembled and tested in-house, with the prototype fabricated on site rather than waiting on an external fab queue.