Output Devices

The Framebuffer Trap

The SSD1306 always operates a 128×64 framebuffer internally. The 64×48 panel shows only a crop of it. Drawing at (0,0) writes to memory the panel never renders.

The full 128×64 framebuffer (red) vs. the 64×48 physical panel (black). Origin (0,0) is completely off-screen. Via Seeed Studio wiki.

The visible region:

X: 31 → 95   (64px wide)
Y: 16 → 63   (48px tall)

Imagine drawing on a 128×64 canvas, then placing a 64×48 window over the centre. Everything outside the window exists in memory — it just isn't shown.

Connection to the Board

The OLED plugs into a 4-pin JST header designed into the board in Week 6: VCC, GND, SDA (pin 8), SCL (pin 9). No soldering after fabrication — the header was already there.

Plug-and-play. The header was designed in KiCad during Week 6 and milled in Week 8.

Failure 1 — Firmware Wasn't Running At All

Upload appeared successful. Nothing happened. Root cause: wrong bootloader configuration. My SWD/CMSIS-DAP workflow does direct flash writes, but the board was set to 4KB_BOOTLOADER — which places firmware at the wrong flash offset. The chip was executing garbage memory.

"Burn Bootloader" with NO_BOOTLOADER is not a contradiction — it's a fuse/clock/layout operation. After this, firmware executed.

Failure 2 — No Way to Confirm Execution

No onboard LED, no USB serial. How do you prove the chip is running? Multimeter on a GPIO pad.

Pin A4 toggling every 5 seconds. Red probe on the pad, black on GND — watching for 3.3V ↔ 0V. No LED required.

Root Cause 1 — Wrong I²C Mapping

OLED blank. I²C scanner found nothing. My board physically routes SDA to A8 (pin 8) and SCL to A9 (pin 9) — but the Fab SAM core was mapping hardware I²C to different pins under the default Serial Config. Fix: switch to NO_UART_ONE_WIRE_ONE_SPI.

Serial Config controls SERCOM-to-peripheral assignment. This one change routes I²C to the correct physical pins. I²C scanner then found the OLED at 0x3C.

Root Cause 2 — Hardware I²C Instability

Scanner worked. OLED still blank. Hardware Wire on ATSAMD11 with Fab SAM core was unreliable with OLED libraries. Fix: switch to software I²C. Bit-bangs the protocol in firmware — slower, but completely stable on this chip.

U8G2_SSD1306_128X64_NONAME_F_SW_I2C u8g2(
  U8G2_R0,
  /* clock=*/ 9,
  /* data=*/  8,
  /* reset=*/ U8X8_PIN_NONE
);

Root Cause 3 — Hidden Framebuffer Crop

OLED finally lit up — text appeared misaligned or invisible. This is the framebuffer crop described above. All drawing coordinates had to be offset into the visible window. Drawing at (0,0) writes to memory the panel never shows.

Root Cause 4 — Flash Overflow

The ATSAMD11 has 16KB flash. Large libraries caused:

FLASH overflowed by 668 bytes

Fixes: smaller fonts, removed sprintf, stripped unused U8g2 features, enabled config.h code size reductions in Build Options, removed #include <Wire.h> (not needed for software I²C). Every byte counted.

Failure 7 — USB Serial Never Worked

USB CDC didn't enumerate with NO_BOOTLOADER on Fab SAM core. Serial Monitor stayed blank the entire project. Fix: used the OLED itself for all debugging — printing raw sensor values and state directly to screen. Faster than USB serial anyway: real hardware, real time, no driver indirection.

Failure 8 — Marquee Flickered with U8X8

Early marquee used U8X8 (the lightweight non-graphics variant) for speed. U8X8 is character-grid based — it scrolls whole character columns, not sub-pixels. Result: chunky, flickering movement. Smooth pixel-level scrolling requires the full U8g2 graphics mode with clearBuffer() + sendBuffer() each frame.

Final Working Setup

Board:          Generic D11C14A (Fab SAM core)
Programmer:     CMSIS-DAP
Bootloader:     NO_BOOTLOADER
Clock:          INTERNAL_USB_CALIBRATED_OSCILLATOR
Build Options:  config.h enabled (code size reductions)
Serial Config:  NO_UART_ONE_WIRE_ONE_SPI
Driver:         U8G2_SSD1306_128X64_NONAME_F_SW_I2C
                clock=9, data=8

The Rendering Loop

Each frame: clear buffer → draw string at current xPos → send buffer → decrement xPos. When text exits the visible window's left edge, advance to the next fact and reset.

u8g2.clearBuffer();

if (touched) {
  const char* message = facts[currentFact];
  u8g2.drawStr(xPos, 40, message); // Y=40 within visible Y:16–63

  xPos -= 14; // 14px per frame × 50fps ≈ 700px/sec

  int textWidth = strlen(message) * 6; // 6x12 font: 6px per char

  if (xPos < -textWidth + 31) { // 31 = left edge of visible window
    currentFact = (currentFact + 1) % factCount;
    xPos = 95;                   // 95 = right edge of visible window
  }
} else {
  u8g2.drawStr(40, 40, "Hold Me");
}

u8g2.sendBuffer();
delay(20); // ~50fps

The numbers 31 and 95 are the visible window edges from the framebuffer crop. Without that knowledge they look arbitrary. With it, they're the exact bounds of what the panel can show.

Final Working Output System

• Display: Grove OLED 0.66" SSD1306, 20×20mm, 64×48 physical pixels
• Driver: U8G2_SSD1306_128X64_NONAME_F_SW_I2C — clock=9, data=8
• Visible window: X: 31→95 · Y: 16→63
• Working features: touch-triggered UI, smooth marquee animation, 9-fact cycle, centered "Hold Me" text using crop offsets