MCU firmware and board bring-up. Debugging notes with code examples and annotated diagrams, backed by sources. Follow for practical hardware checks.

4.7 kΩ + 100 pF gives a calculated I2C rise time of 398 ns. Fast-mode (400 kHz) allows 300 ns. Check both rise time and sink current before choosing pull-ups. TI: ti.com/lit/pdf/slva689
1
7
382
On Cortex-M7, invalidating a cacheable 16-byte DMA RX buffer can also discard dirty data sharing its 32-byte cache line. Reserve whole lines for RX: align the allocation start and pad it. ST's buffer guidance: github.com/STMicroelectronic…
23
24LC256 can ACK a write command with WP high and leave the EEPROM unchanged. Check WP at STOP. Once the device is ready, read back the target address and compare the bytes before reporting success. Microchip DS20001203Y sections 6-7: ww1.microchip.com/downloads/…
2
33
STM32F411 USART2, PCLK1=16 MHz, OVER8=0, BRR=0x008B: the calculated rate is 115,108 baud, about -0.08% from 115,200. Read back BRR and check PCLK1. This is divider error only; oscillator error is extra. ST RM0383 Rev 4 §19.3.4: st.com/resource/en/reference…
2
77
STM32F4 RX with double-buffer DMA: CT=0 writes M0; CT=1 writes M1. Process the other buffer before DMA switches back. Example: 1,024 samples at 48 ksample/s leaves ~21.3 ms for interrupt latency and processing. ST AN4031 §1.1.11: st.com/resource/en/applicati…
65
This UNO Q ADS-B guide checks readsb's RTL-SDR build support before connecting its Beast output to fr24feed on port 30005. A dongle showing up in lsusb is only the USB-layer check. rs-online.com/designspark/fe…
105
Before wiring a mystery seven-segment display, map its pins. Here the author found separate red and green anodes on pins 3 and 8. Photo: @kumanopooh333.
秋月の2026年福箱に入っていた7セグLED。型番「SD-AG106URG」でググる。前年以前にも福箱行きのパーツで緑色のアノードコモンらしい。念のため動作確認すると一般的な7セグと若干ピン配置が異なる。またアノード端子が共通でなく、3ピンが赤色・8ピンが緑色の2色LEDだった。だから末尾がRGか。
2
80
Sebastian's first diagonal split into two segments because both axes stepped at the same rate. The fix was a Bresenham-style step schedule, shown at 17:59 in the video. Photo: Sebastian Lague / Microchip Makes.
Sebastian Lague's custom-built, ATmega4809-controlled pen plotter draws unique topographic maps of the Moon, Mars, and Earth: piped.video/qyNWDzO6kKQ
45
wolfBoot reports ~31 ms to load a 180 KB signed image from SD on the LS1028ARDB, but ~1.45 s from card initialization to a verified image. Card power-up dominates that measurement. I'd include it in a boot-time budget. wolfssl.com/wolfboot-sd-card…
1
40
Cameron Coward's Termi3 runs Llama 3.2-1B on the UNO Q's Linux side. The STM32 handles the Silent 700's 300-baud, 7-bit, odd-parity link through a MAX3232. The build includes both Python and MCU code.
An Arduino UNO Q lets this vintage terminal answer questions without internet access. blog.arduino.cc/2026/10/01/a…
41
AMD is sampling its enhanced XQRVC1902 package. Class Y flight-qualified parts are expected in H2 2027. For board reuse, I'd check the lidless package's thermal interface even with the same 2197-ball pinout. Source: newsroom.amd.com/news/amd-sa…
37
Raspberry Pi's Oct 1 update puts the 2GB Pi 4 at $67.50 and 2GB Pi 5 at $77.50, both up $12.50. Before choosing a smaller RAM option for an embedded build, I'd measure peak use during boot and software updates.
Due to the continuing rise in the cost of memory, we’ve had to raise the prices of our 2GB Raspberry Pi 4 and Raspberry Pi 5 products. Eben explains the cost increase – along with ways to ensure you’re not paying for more memory and processing power than you need – in today’s blog post. raspberrypi.com/news/price-i…
44
In KB Sriram's 30-card ATtiny sieve, REQ-to-VOTE low measured about 950 ns, while settled high took 1.684 µs. The release edge includes the shared line charging. I'd measure both edges before setting a sampling deadline. Build log: hackaday.io/project/206556-a…
1
35
One detail I'd keep with a failed-chip report from this Pico 2 tester: the measured 5 V rail. The README warns that low USB voltage can make a working DRAM fail its speed grade. Project: github.com/schlae/pico-dram-… Photo: Tube Time / schlae
New blog post: Raspberry Pi Pico 2 runs retro RAM chip checker. blog.pimoroni.com/raspberry-…
2
43
In PULP's AXI NoC study, the credit-based router needed three FIFO slots to hide credit-return latency. Two slots cut sustained link utilization to 66% in the tested setup. I'd check that latency before choosing the buffer depth. Paper: arxiv.org/abs/2607.01430
38
Why can six EEPROM bytes need two page writes? On an AT24C256C, address 62 leaves 2 bytes in a 64-byte page. Split 2 + 4, waiting for each write cycle to finish. Illustrated example; no hardware test. Free walkthrough: patreon.com/ManBoEmbedded/po…
24
The $30 price is for the carrier; CM5 is sold separately. It can be reflashed in place through the USB-C SERVICE port, with the display supplying power. I'd leave access to that port and the boot switch when designing a kiosk enclosure. Docs: raspberrypi.com/documentatio… Photo: @cnxsoft
Raspberry Pi CM5 carrier board for digital signage and kiosk applications. cnx-software.com/2026/10/01/… The @Raspberry_Pi Smart Display Module is compatible with the Intel Smart Display Module (SDM) specification, more specifically the SDM-L form factor, and designed to be plugged into compatible displays The $30 board exposes a standard PCIe x8 SDM connector with HDMI 2.0, USB 3.0, I2C/SPI/UART, and power signals, and features a GbE RJ45 port, a full-size HDMI connector for an external display, a USB 3.0 port, and more.
27