MR1 Nanopore Reader

The repository began on November 11, 2025, with a commit whose message is "first commit". Six weeks later, on December 23, a commit added "Sentinel Nanopore Reader Rev 11" as a file for OpenSCAD, a program that builds 3D models from scripts.^[9][10] Over the following year the mechanical design moved to build123d, a Python library for parametric CAD, and the electronics were drawn in KiCad. A board for the pump and its power supply was added in September 2026 under the name "Sentinel Pump & USB-PD Hub".^[1]

The design is led by the company's chief systems architect, whom it describes as a "master hardware prototyper".^[5] Of the 454 commits in the repository by October 20, 2026, 143 were made by the bot account of an AI coding agent.^[1][4]

On October 15, 2026, a commit prepared the repository for release. In its own words it "Adds AGPLv3 license to all relevant source files and root directory, cleans up temporary files and cache directories, updates README and documentation, standardizes electrode terminology, and temporarily skips specific KiCad build verification tests until CI support is available."^[11] The citation file gives the title "An Open-Source Low-Noise Solid-State and Biological Nanopore Sensing Platform", the version 1.0.0 and the release date October 16.^[2] At release the README opened with the name "Sentinel Nanopore Reader (Daemon Pore)", and its test badge reported 95 tests passing.^[4]

Two days later a commit titled "refactor: Rename Daemon Pore/Sentinel to MR1 across codebase" changed 71 files, adding 139 lines and deleting 139.^[6][8] The name MR1 had until then belonged to the reader board inside the stack and to its firmware folder.^[4] At least one line escaped the change: as of October 25, a design document still gave its "System Target" as "MR1 Nanopore Reader (Daemon Pore)".^[1] The four-byte word that marks the start of every data frame the reader sends is "SENT".^[a]

Three days after the rename, on October 21, a commit titled "Relicense project to Public Domain under CC0 1.0 Universal (CC0-1.0)" placed the whole project in the public domain.^[3]

From the top, the assembled design has a torque cap, the cartridge, the reader housing with the main circuit board, and the pump module. The cap screws down on a thread 75 mm across and is fitted with a ratchet that slips at 0.5 newton meters, so that every cartridge is clamped with the same force.^[1][5] Five lines of tubing and three bundles of wiring join the modules, and four Luer ports on the lid of the reader take fluid in and out.

The cartridge is about 22 mm tall. In order from the top it holds a pressure puck with a hole for a pipette tip, a hexagonal plate of polypropylene, a cast silicone gasket (PDMS) holding the chip, a circuit board of PTFE or Rogers laminate called the interposer, a second gasket, a lower hexagonal plate, and a pusher plate of TPU.^[5] The chip is a die of silicon 4 by 4 mm and 500 µm thick, turned 45 degrees in its gasket, with a thin window of silicon nitride in the middle. Its bill of materials lists membranes 10 to 50 nm thick with pores of 2 to 10 nm; the company's worked example is a 100 nm pore in a region thinned to 200 nm.^[5]

The window is made by etching the silicon from the back with hot potassium hydroxide, which eats into silicon until it reaches the slow-etching {111} crystal planes and leaves walls sloped at 54.74 degrees. A window 10 µm wide on the front of a 500 µm die therefore needs an opening about 717 µm wide on the back.^[5]

Two peristaltic pumps move salt water through the chambers above and below the membrane, called cis and trans. Each is driven by a DRV8837C motor driver rated at 1 A, and two pressure sensors watch the feed line (0 to 15 psi) and the trans line (0 to 5 psi); the pump firmware stops at 10 psi and 4 psi respectively, and limits the flow to between 0.5 and 150 µL per minute.^[5] The tubing is PharMed inside the pump heads, FEP on the feed line, PEEK on the trans line and silicone for waste. In channels this small the flow is laminar, with a Reynolds number of order 1 to 10.^[5][b]

The membrane is the fragile part. A film tens of nanometers thick can tear if the pressure on one side runs ahead of the other, and the design balances the hydraulic resistance of the cis and trans paths for that reason.^[1]

The interposer carries four silver chloride electrodes, ribbons of silver 2.0 by 0.8 by 0.1 mm chloridized in 0.1 M ferric chloride before assembly. On the cis side a working electrode drives the current and a sense electrode measures the potential; on the trans side a reference electrode measures and a counter electrode returns the current. The sense electrodes sit about 6.9 mm from the pore and the drive electrodes about 15.8 mm away.^[5] The aim of the four-electrode cell, according to the design documents, is to keep the polarization of the electrodes out of the voltage that is measured across the pore. That is a statement of intent; nothing has been measured.^[5]

The voltage across the membrane comes from an MCP4822, a two-channel 12-bit digital-to-analog converter, combined with a 2.048 V reference. One channel sets the bias in steps of about 1.2 mV and the other trims it in steps of about 12 µV, over a range of about ±2.46 V, as drawn.^[5]

The current is read by a transimpedance stage built on an LMP7721 operational amplifier, whose input bias current is 3 fA typical and 20 fA at most.^[12] Three femtoamperes is about 19,000 electrons a second, one for every million or so ions that cross a pore carrying 3.1 nA. The feedback network is 1 MΩ in parallel with 2 pF, which sets a pole at 79.6 kHz. A second stage on an LTC6268-10 multiplies the signal by ten, a fully differential driver feeds it to a 16-bit or 18-bit successive-approximation converter, and the overall gain is about 10 MΩ: one volt out for every 100 nA in.^[5] Depending on the converter's reference, the full scale is between about ±240 nA and ±510 nA. On October 22, 2026, a second, wideband channel was added to the amplifier schematic for much larger pores, built on an OPA192 operational amplifier with a 100 kΩ feedback resistor; its document gives it a span of ±20.48 µA, for apertures of 10 and 100 µm.^[1]

Those figures suit pores from a few nanometers to about 100 nm across. In 1 M potassium chloride at 200 mV, a 10 nm pore in a 20 nm membrane would carry about 5.9 nA, and the company's 100 nm example about 59 nA. A 2 nm pore, at about 0.3 nA, would sit roughly 20 times above the noise in a 10 kHz band, and a pore 1 µm wide would overrange the converter.^[5] The thermal noise of the 1 MΩ feedback resistor alone is about 12.8 pA RMS in a 10 kHz band, and no amplifier built around that resistor can be quieter.^[5]

Design documents in the repository describe a more sensitive front end than the schematic draws:

Everything runs from one USB-C cable. On the pump board an STUSB4500 negotiates between 9 and 20 V from a USB Power Delivery charger, a TPS54302 converter (rated for inputs of 4.5 to 28 V at 3 A) brings it down to 5 V, and a TS3USB221 switch passes the reader's data through to the host computer. An STM32G431 microcontroller runs the pumps.^[5] In the reader, the analog supplies pass through a charge pump, transistor filters and a pair of low-noise regulators, the LT3045 and LT3094, which add about 0.8 µV RMS between 10 Hz and 100 kHz. The converter's own 1.8 V supply comes from a separate regulator whose enable pin doubles as the leak interlock.^[5]

The ESP32-S3 has two cores. One reads the converter and streams the samples over USB; the other runs Wi-Fi, a web oscilloscope that shows the trace decimated to 1,000 points a second, and the link to the pump module.^[1] Samples travel in frames of 146 bytes, each holding the four-byte sync word, a frame number, a timestamp in microseconds, a sample count, status flags, 64 samples of 16 bits and a CRC-16 checksum. At 100,000 samples a second that is 1,562.5 frames a second, about 1.83 Mbit/s, which fits within full-speed USB at 12 Mbit/s.^[5] Saved at the same rate, the raw trace fills about 720 MB an hour. A Python client on the host computer reads the frames and has a mock mode that invents baseline current and translocation events for testing.^[1] The firmware can also apply sine, triangle, square, ramp and chirp waveforms to the bias, and its default pulse for clearing a clogged pore is 800 mV for 250 ms.^[5]

Salt water above an amplifier sensitive to femtoamperes calls for a leak alarm. The design surrounds the wet area of the interposer with a trace, the weep moat, laid 0.25 mm from a copper ground fill. Dry, the gap measures more than 10 GΩ; bridged by salt water, it falls below 1 kΩ. In the design, a breach switches off the regulator that powers the converter, and the firmware then stops the pumps and sets the bias to zero.^[5]

The repository's test suite checks the design for consistency. Six suites look for collisions between the 3D models of parts, from the cartridge plates to the tubing routes; others assert the dimensions of every printed and milled part against reference models, check that the 75 mm cap thread engages without overlap, and compute the curvature along every tube and wire to confirm it never bends tighter than 10 mm. Firmware tests check the frame format, the checksums and the mapping from voltage to converter code, and an end-to-end test confirms that each of the 159 files the build exports exists and matches its recorded SHA-256 hash.^[1] None of the tests involves an electrical measurement.