pic-firmware

PIC & dsPIC firmware (MPLAB X + XC + MCC; PICkit / Snap / ICD)

Also load embedded-c-cpp for coding standards, architecture, unit testing and general debugging. This skill holds the Microchip-specific parts. PIC firmware is written in C (XC8 has no C++).

First, identify the project

Find	Where	Why it matters
Exact part (PIC16F18877, PIC18F47Q10, PIC24FJ128GA010, dsPIC33CK256MP508, PIC32MX795F512L...)	nbproject/configurations.xml (<targetDevice>), MCC .mc3, user	Registers, peripherals, PPS, interrupt model, memory all differ per part — on dsPIC33 the F, E, C and A families have different PWM/ADC/PLL modules
Compiler + version	configurations.xml (<languageToolchain>, <languageToolchainVersion>)	XC8 vs XC16 vs XC-DSC (dsPIC) vs XC32; optimization levels and syntax differ
MCC flavour	mcc_generated_files/ layout, .mc3 file; Melody (newer: system/, timer/ subfolders, *_Initialize) vs Classic	Where config bits and drivers live, API names
Clock	MCC clock settings / #pragma config (FEXTOSC, RSTOSC / FNOSC / POSCMOD), _XTAL_FREQ or FCY define	Delays, baud rates, timer periods
Programmer/debugger	PICkit 4/5, MPLAB Snap, ICD 4/5	Power and high-voltage programming capability

When a register or bit name is uncertain, say which datasheet section to check (or read the device header pic16f18877.h etc. in the compiler's include directory) rather than guessing — PIC register names vary between sub-families.

Rules that prevent the classic PIC failures

1. Digital input reads 0 → ANSEL. On most modern PICs, pins power up as analog. Clear the ANSELx bit (MCC: uncheck "Analog") for every digital input.
2. Write LATx, read PORTx. Writing PORTx bits does read-modify-write of the pin levels and can flip other outputs (heavy loads, fast toggling). Parts without LAT registers need a shadow variable.
3. TRIS: 1 = input, 0 = output. Easy to invert.
4. Peripheral Pin Select (PPS) must map UART/SPI/PWM signals to pins on PPS devices; with PPS1WAY = ON it can be unlocked only once after reset — let MCC generate it.
5. Config bits decide oscillator, watchdog, MCLR, LVP, brown-out, code protection. Set them in MCC (or the Configuration Bits window → Generate Source Code), never leave them to defaults you did not check. Wrong oscillator = nothing runs or every timing is off.
6. _XTAL_FREQ (XC8) / FCY (XC16) must equal the real clock — __delay_ms() and many calculations depend on it.
7. 8-bit atomicity: on PIC16/18 any uint16_t/uint32_t shared with an ISR is read in several instructions — disable that interrupt (or GIE) around the read. On PIC24/dsPIC the same applies to 32-bit values.
8. Hardware call stack is tiny on 8-bit PICs (8 levels on old PIC16, 16 on enhanced PIC16F1, 31 on PIC18) and the ISR uses levels too. No recursion; watch XC8 warning (1393) possible hardware stack overflow and the estimated depth in the build output.
9. Interrupt flags: in the ISR, check both enable and flag bits (if (PIEx.xIE && PIRx.xIF)), clear the flag, keep it short. Enable peripheral + global interrupts (MCC: INTERRUPT_GlobalInterruptEnable()).
10. MCC regeneration overwrites mcc_generated_files/. Put application code in your own .c/.h files; keep main.c thin; review the MCC merge dialog instead of accepting blindly.
11. Irreversible / risky settings: code protection (CP), write protection, disabling MCLR together with LVP off, oscillator settings that stop the chip. Don't enable code protection or change these on a user's device without explicit confirmation; a Snap cannot recover a part that needs high-voltage programming.

Family-specific syntax (interrupts, delays, config, memory): references/families.md.

dsPIC33 (control, DSP, motor control, digital power)

dsPIC33 uses the PIC24 core syntax plus a DSP engine. Extra rules:

• FCY = FOSC/2; PLL setup and clock switching differ per family (EP vs CK/CH) — generate with MCC and verify the reported frequencies, including the auxiliary clock that drives PWM/ADC.
• Use Q15 fixed-point (DSP library dsp.h, libq.h, builtins) instead of float in fast loops; document every variable's scaling.
• Clamp integrators and outputs explicitly in C — normal C arithmetic wraps; CORCON saturation only protects DSP-engine operations.
• Control loop = PWM-triggered ADC interrupt at high priority; measure its execution time on a scope.
• Configure PWM fault/current-limit inputs before enabling outputs on any power stage, and test dead time at low voltage first — wrong PWM code destroys hardware.
• Enable the catastrophic accumulator-overflow trap (COVTE); leave OVATE/OVBTE off for normal MAC code (guard bits are meant to be used).
• dsPIC33CK/CH alternate register sets use __attribute__((interrupt, context)) with matching ISR priority — not shadow.
• dsPIC33CH is dual-core: two projects, pin ownership, MSI mailboxes.

Details: references/dspic33.md.

Build, program, debug

Action	How
Build in IDE	MPLAB X → Clean and Build (hammer+broom). Check the Memory Summary and warnings in the output
Build headless	prjMakefilesGenerator.sh <proj>.X@default then make -C <proj>.X CONF=default → dist/default/production/<proj>.X.production.hex
Program	MPLAB X "Make and Program Device", MPLAB IPE, or ipecmd from the command line
Debug	MPLAB X Debug Main Project with PICkit/Snap/ICD (debug build), or the MPLAB X Simulator for pure logic/timers without hardware

Programming changes the device: confirm the part and programmer with the user before running ipecmd, and never erase/protect without confirmation. Full commands, programmer differences and connection troubleshooting: references/build-program-debug.md.

When it doesn't work

• Nothing runs: config bits (oscillator source, WDT, MCLR), power, MCLR pull-up (10 kΩ), programmer released the device from reset (IPE "Release from Reset" / -OL).
• Runs at wrong speed / wrong baud: RSTOSC/FNOSC, PLL, clock divider (CDIV), _XTAL_FREQ mismatch.
• Random resets: read the reset-cause bits at boot (PCON/PCON0/RCON — references/families.md), watchdog not cleared, stack overflow (STKOVF), brown-out.
• PIC24/dsPIC sudden reset or hang: install trap handlers (_AddressError, _StackError, _MathError, _OscillatorFail) — address errors usually mean a misaligned 16-bit access.

Reference files

File	Read when
references/families.md	Writing/reviewing code for a specific family: interrupt syntax, delays, config bits, memory model, atomicity, reset causes, traps, PIC32 cache/DMA, XC8 optimization and memory tips
references/dspic33.md	Any dsPIC33 work: families, XC-DSC, PLL/clock switching, DSP engine and Q15 math, ISR context/shadow registers, math traps, DMA, high-speed PWM/ADC control loops, dual-core CH, motorBench/power tools, review checklist
references/mcc-and-project.md	Project structure, MCC Melody vs Classic, regeneration rules, where to put code, peripheral drivers (UART, I2C, ADC, timers, PWM, NVM/EEPROM), low power
references/build-program-debug.md	Installing tools, headless builds, ipecmd programming, PICkit/Snap/ICD differences, debugging limits, "cannot connect / device ID mismatch" troubleshooting, unit testing PIC code on the PC