Flow · Lattice OS
µF

Microfluidics

Whole-array microfluidic control that runs locally.
native Rust controllercloud-independentexact replay

Large active matrices are usually controlled as collections of independent channels. Flow carries the active control state through the complete field instead. The worked example below is the native Rust controller produced by that architecture: it authorizes, checks, routes, and records each exact control action locally, without putting a cloud service inside the control loop.

WORKED EXAMPLE   A registered 256×256 active matrix with 65,536 independently addressable sites.
EDGE RESULT   A standalone Rust controller completed the same 20,000-route workload in 15.4 milliseconds.

A working controller at the edge

Bench-tested Rust control path

20,000 control decisions in 15.4 ms

Time to complete the same worked control cycle. Less is better.

Each clean-path route builds and seals its packet, verifies the packet and controller image, derives the addressed finite-field word, senses the syndrome, applies the exact action, and returns the routed output. The measurement ends at the software controller output. Exact readings: public evidence receipt.
Cloud-independent operation

The control loop stays with the machine.

The compiled controller image travels with the edge runtime. Once deployed, sensing, authorization, repair, projection, recursive action, and receipt generation execute locally. A network may deliver a new approved image or collect receipts, but the running controller does not require a live connection.

200 KiBstandalone executable
16,513 Bsealed controller image
3.84 MiBmeasured working set
85 Bsealed command packet

Execution dependencies: Rust standard library only. No Python, NumPy, network service, cloud runtime, or third-party crate.

software propertydelivered result
registered active-matrix fixture256×256; every one of 65,536 addresses encoded, decoded, and checked
address representation26 compatible coordinate words plus one action word
controller implementationstandalone native Rust executable
edge image16,513 bytes
command packet85 bytes
fixed control state936 bytes
measured process working set3.84 MiB on the registered host fixture
recursion replaydepths 1–81 byte-identical to the independent reference
execution dependenciesstandalone Rust; no Python, NumPy, network, cloud, or third-party crate

Why this worked example matters

The Flow scaling benchmark establishes that recursive full-field transport does not turn a larger address space into a larger control scan. This microfluidic example establishes the other half of the product claim: the same architecture has been reduced to a small, fast, native controller that runs without a cloud connection. The 256×256 fixture is the exhaustive correctness test, not the limit of the logical field.