ZFS pool went degraded… not one, but all 4 NVMe’s reporting SMART errors with 4% of their 600TB write endurance used. TRIM was enabled on the pool. They were mounted to a heatsink with active cooling on an x16 bifurcated carrier and ran cool. Warranty expired 6 days ago. I sent SanDisk logs and smartctl output and thankfully they made an exception to RMA all 4. I don’t even want to look at what 1TB NVMe’s are going for these days…

Missed some holes and didn’t realize until I already reflowed.. back on the CNC with the same registration pins

At the grand slam, watching Canada vs South Korea women’s finals

Relief carving some wood on the Shapeoko CNC

I'm increasingly convinced high purity quartz (HPQ) may be part of the strategic interest in Greenland and Canada. HPQ is critical to semiconductor manufacturing, particularly the crucibles used to grow silicon crystals. Global supply is overwhelmingly concentrated in Spruce Pine, North Carolina, where detailed reserve figures are closely guarded. How much is left? Good luck finding out. What do Greenland and Canada have in common? Quartz. Semiconductor crucibles require exceptional purity, typically 4N–5N (99.99–99.999% SiO2), with strict impurity limits. Greenland has ~6 million tonnes of estimated quartz at Ivittuut, with samples reaching ~3N purity. But with 80% of the island buried under ice and much of its geology underexplored, that may be just the tip of the iceberg. Its true HPQ potential remains largely unknown. In Canada it gets interesting. BC: Deposits at Mt. Rose and Armstrong have reportedly been successfully refined to 4N8 purity. Manitoba: An estimated 15 BILLION tonnes of silica in situ. Sio Silica reports approximately 540 million tonnes potentially recoverable, describes it as the world's largest deposit, and claims laboratory processing has achieved 5N purity. If even a small fraction of Canada's enormous silica resources can economically produce crucible grade HPQ, it's of extreme strategic value, especially given the unknown remaining reserves at Spruce Pine.

Reflowing some Infineon AS5047U magnetic position sensors to try out. I’ve used the D variant before which work very well, but high res needed SPI readout. These have SPI for config and angle readout plus PWM, UVW, and ABI outputs but this U variant has full 14 bit, 16384 count resolution on ABI. ABI is much more deterministic, lower latency with less overhead on an MCU with timers that support quadrature encoders in hardware, since you just need a single cycle read or a DMA trigger to get the angle. They also have angular velocity estimators, averaging filters, and dynamic angle correction DSPs that compensate for internal delays vs angular velocity

Next going to try this RC Novak 540 10.5T motor. It has hall sensors which I’ll hookup to the JST hall input of the board later, trying senseless BEMF first

The firmware is built around deterministic runtime graphs. Can switch between sensor feedback, sensorless control and different observers over CAN. Shared STM32 drivers support HRTIM and advanced timers; BSPs provide board mappings. Reconfigure, tune and save without reflashing.

Finished bringup of the STM32 G474 LMG5200 half bridge GaN 3 phase BLDC controller board. Testing using a small gimbal motor in FOC with sensorless BEMF observer feedback

"Compiling rust is slow"

My custom PCB CAM is running its first live job on the CNC.. Workflow is directly load a kicad PCB, panelize N boards, click generate, and run. It automatically selects the best tools from persistent ATC magazine loadout and generates toolpaths using available tools. It picks tools appropriate for isolation and will do rest machining only using the smallest tool where necessary. Picks endmills for routing, does inside contours for holes, outside contours for edge cuts. Ensures climb milling direction on material engagement. It is net aware and verifies the nets are isolated, generates an NC, uploads it to the machine, runs Z zero and 9x9 mesh compensation over the NC bounds, and runs the program. This brings 5-10 minutes of manual CAM setup and visual inspection of manufacturability (ensuring tools actually fit to isolate copper) down to seconds.

Surface mount populated

Finished up on the CNC.. time for paste and reflow on the @voltera_io

I still have a handful of the TI LMG5200 GaN half bridges on hand. They’ve been working great on my other boards so I decided to layout a 3 phase BLDC controller for my custom rust FOC stack. This will use a STM32 G474CET, with complementary HRTIM channels to three LMG5200s, a small buck for 5V, two bidirectional INA254’s inline with two of the motor phases for current feedback, and some JST connectors for either hall sensors or SPI/ABI magnetic encoders. Molex microfit 3.0 for power, and surface mount M4 screw terminals for the motor. And of course a CAN FD transceiver for C2 I just got some updated AS5047U magnetic sensors that have 4096 pulse/rev quadrature output that I’ll be trying out.. I’m excited for this since you used to have to get 12 bit angle over SPI, and ABI output was decimated to 10 bit on the previous revisions. ABI is connected to a hardware timer in encoder mode, so getting shaft angle is just a register read rather than having to schedule SPI transactions aligned with PWM.

Added multiple PCB support in the CNC CAM app I've been working on. Loads native kicad_brd or gerber job manifests, auto arranges them onto defined stock and can export excellon and paste gerber for the whole panel so I can run multiple small boards at once on the Voltera