What is it about?

This board is the first piece of hardware for AI MotoLogger, a data logger I'm designing for a KTM 250 SX-F motocross bike. The final logger will record motion (IMU), position and speed from a dual-frequency GNSS receiver, suspension travel, brakes, wheel speeds, engine data from the ECU's CAN bus (listen-only) and a wideband lambda sensor. The target is to time-stamp every channel against GNSS time to within 100 µs.

Before designing the main board, I have to choose its IMU. That's what this board is for: it puts four IMUs side by side on one rigid 4-layer PCB, so they all feel exactly the same shocks and the same vibration.

Why did I make it?

A motocross bike is a hard place for an inertial sensor. Landings, ruts and crashes produce acceleration peaks that I haven't found measured anywhere, and most IMUs stop at ±16 g. The engine also shakes everything, and that vibration can shift a gyroscope's offset. Datasheets rarely say how a part behaves in that environment.

Four separate breakout boards wouldn't answer the question: each sensor would sit on a different board, with its own mounting and its own resonances. On one rigid PCB the comparison is direct, and it costs less than four evaluation modules.

The four candidates:

TDK ICM-45686: ±32 g, ±4000 °/s. The current choice for the main board.

ST LSM6DSV320X: ±16 g plus a high-g channel up to ±320 g in the same FIFO. If it performs well, one chip could do the job of two.

ST ASM330LHH: automotive grade (AEC-Q100), rated to +105 °C, but only ±16 g. The backup if the enclosure gets too hot.

Analog Devices ADXL375: a ±200 g accelerometer, used as the reference for peaks above 32 g.

How does it work?

Board: 54 × 50 mm, 4 layers (signals / GND / 3.3 V / signals), 1.6 mm thick, all parts on the top side. It has four M3 mounting holes and two extra holes for a cable tie next to the connector.

Power: 5 V comes in from the development board through a ferrite bead. An on-board 3.3 V LDO (Torex XC6206) feeds the sensors, and every power pin has its own decoupling capacitor.

One SPI bus for all four: SPI mode 3 at 5 MHz, with a separate chip select for each sensor. The ADXL375 can't switch off its I²C interface, so SPI traffic meant for the other sensors could look like I²C to it. A single-gate OR (SN74LVC1G32) in front of each sensor's data input fixes this: while a sensor isn't selected, its input is held high.

Hardware time stamps: each sensor's interrupt line goes to a 32-bit timer input-capture channel on an STM32U5 Nucleo (NUCLEO-U5A5ZJ-Q). Every FIFO batch gets a precise time stamp, with no interrupt latency.

One connector: a 1×15 header with 2.54 mm pitch carries 5 V, GND, SPI, four chip selects and four interrupts. On the bench it uses Dupont wires. On the bike, the cable is soldered and tied down.

MEMS-friendly layout (following ST's TN0018):

No copper, tracks or vias under the sensors on the top layer.

The sensors sit in a row with their axes aligned (the ADXL375 is rotated 90°), and the axes are marked on the silkscreen.

Stencil apertures are reduced on the LGA pads, so there's less solder paste under the sensors.

A small check for the AliExpress part: the ADXL375 reports the same device ID as the ADXL345. The firmware therefore checks the reading at rest: about 20 counts per g for a real ADXL375, about 256 for a relabelled ADXL345.

Test plan:

On the bench: check that all four respond, measure noise at rest against the datasheets, and hit the board with a rubber mallet to cross-check the high-g channels.

On a Yamaha YZ 125 test bike: find out whether real peaks go past 16 g or 32 g, record the engine vibration spectrum and gyro drift under vibration, and read the board temperature from each sensor's internal thermometer.

The results will decide the main board's IMU: the LSM6DSV320X alone, the ICM-45686 alone, or one of them plus the ADXL375.

Status

I'll assemble the board by hand using the stencil, then test it on the bench and on the bike. I plan to update this page with the results.