A pocket Korg M1 revival on the RP2350

Nov 05, 2026 Last reply: 20 hours ago 2 Replies
Jump to replies (2)

The whole thing started as a want: the maker felt the famous Korg M1 belonged in the PicoVintageSynthCollection, a growing set of vintage synth recreations, and set out to build it. The result is PicoFaceM1, a DIY synth running on a stock RP2350 board with 8 MB or more of flash, wearing a green PCB and a small color screen, with three rotary knobs marked SW1, SW2 and SW3.



What stands out to me is how faithful the engine is. It was written from scratch by disassembling the original M1 operating system, with no borrowed code. The filter path mirrors the real hardware with four single-pole sections split across two chips, hitting a 24 dB/oct slope from the genuine tables. Amplitude is handled multiplicatively, with linear ramps stepping every 20.8 ms. Every one of the 33 original effect types came over as an algorithm, and it holds 16-voice polyphony. Samples stream straight forward out of 4 MB of PCM in flash.



To keep audio blocks on schedule, the core runs overclocked at 444 MHz and 1.60 V. That seems a sensible budget to me, and the maker has said the true bottleneck is not the clock speed itself but pulling samples out of slow flash quickly enough, which takes block transfers.



The demo runs through patch presets and combination banks while playing audio, with names like Universe, Piano 16', Brass 1, Solo Sax, Drums #1 and FilmScore appearing on screen, and the unit reproduces both the synth sounds and multi-timbral sequences convincingly. One thing still unclear is the exact board sitting under the assembly. Note that users must supply their own legally obtained PCM data and OS ROMs, which the build system picks up automatically from the roms/ directory.


The filter recreation is the part I keep coming back to. Splitting four single-pole sections across two chips to copy the original circuit, rather than simplifying it into one generic filter block, is a clever call. It costs more processing headroom, which is likely part of why the 444 MHz clock budget exists, but it should preserve the character of the slope far better than an approximation would.

An idea for the maker: the extraction process has only been checked with Rev 19 firmware so far. It would be useful to confirm the same reading procedure works on other OS revisions too, since owners of the real instrument may hold different versions in their machines. That would widen who can actually use this, given the ROMs and PCM data always stay user-supplied and out of the repository.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required