Layer1Labs is raising seed funding — $1–1.5M on a SAFE, seeking a lead investor. Talk to us →
Hardware

ChronoFabric

The hardware tier of the same architecture.

ChronoFabric takes the programs that run in the ChronoHive software engine and enforces them on FPGA and hardware endpoints — the same deterministic guarantees, delivered at hardware speed. ChronoFabric is available now as soft IP — a licensable hardware IP block — targeting Kria FPGA platforms. Embedded chiplets and merchant silicon are future offerings on the staged hardware roadmap.

Complete and ready for beta customers

Same Programs, Hardware Enforcement

Programs are authored once, in Lingua Franca, and compiled by chronoc into a single constraint specification (.cspec). That specification is the deployment contract for both tiers: the ChronoHive engine runs it in software, and ChronoFabric runs the identical specification on hardware.

Only the effector binding changes. An effector is a connection point where a program acts on the outside world — a storage write, a network send, a hardware action. In software, effectors bind to software callables (ordinary program functions). On ChronoFabric, the same effectors bind to FPGA and hardware endpoints (the physical interfaces of the device).

Nothing else about the program changes: the engine core’s admission (which work is allowed to proceed), sequencing (the order work happens), and determinism guarantees carry over unchanged. ChronoFabric is delivered today as soft IP — a licensable hardware IP block — targeting Kria FPGA platforms, with embedded chiplets and merchant silicon as future offerings on the staged roadmap.

Why Hardware

Software coordination has a ceiling: every decision costs processor time on the critical path. Enforcing the same deterministic schedules in hardware removes that overhead. ChronoFabric is a deterministic orchestration architecture designed to unlock additional AI compute capacity by reducing synchronization waste across large-scale distributed accelerator systems.

The outcome classes are the ones our partners already plan around:

  • Throughput recovery — less idle time across the fleet.
  • Power avoidance — smoother infrastructure-level energy demand.
  • CapEx efficiency — more useful work per watt, and per facility.

Measured on Hardware

Measured on Kria FPGA evaluation boards (KR260/KV260) over 10-gigabit links — the beta platform. Production-scale fabrics (100G/400G, ASIC) remain the roadmap, not a measurement.

Measured on hardware: in validation runs between two Kria FPGA boards over a 10-gigabit Ethernet link, ChronoFabric delivered more than 21,000 scheduled packets with zero missed departures and zero dropped — a 100% delivery rate, including a continuous 10,000-packet endurance run.

Packets leave at the exact programmed tick: across 2,000-packet runs, the median and 99th percentile departure interval were exactly the programmed 1 ms, with jitter (standard deviation) as low as 61 nanoseconds — roughly 160× tighter than the same boards scheduling from the CPU.

Saturation-proof: with all four CPU cores held at 100% load, ChronoFabric's scheduled departures still landed on the programmed tick (median and p99 exact), while CPU-scheduled departures on the same boards degraded to millisecond-scale jitter. The hardware scheduler issues no per-packet work to the CPU at all.

Holds under a traffic flood, too: under a flood of 50,000 competing packets, the scheduled flow still departed on its programmed 1-millisecond tick at the median and at the 99th percentile, hitting its target for 99.82% of packets.

In a 10,048-packet run, the fabric's adaptive scheduler adjusted its own timing 53 times in response to on-chip constraint signals — and still delivered every packet: 10,048 of 10,048, zero misses.

Sustained, not just sampled: over a full 24-hour run on a Kria KV260 board — 864 million timing cycles — ChronoFabric's real-time software stack (PREEMPT_RT kernel with isolated CPUs) held worst-case scheduling latency to 22 microseconds on the isolated core, with 99.99% of cycles at or under 12 microseconds.

All three configurations, a full day each: the 24-hour run was completed for all three software configurations on a Kria KV260 board — 864 million timing cycles each. The full hardware profile passed with a worst case of 22 microseconds (99.99% of cycles at or under 12 microseconds). The real-time software configuration on its own also passed — worst case 33 microseconds, 99.99% at or under 15 microseconds. The stock configuration, measured for contrast, reached a worst case of 904 microseconds (99.99% at or under 25 microseconds) — the gap the real-time configurations close.

Across power states: timing held with the board’s processors in performance, power-saving, and partially-offline states, and while cycling under stress — scheduled departures hit their target in 99.94% to 100% of packets, with median and 99th-percentile intervals exactly the programmed 1 ms in every state.

Two boards, one tick: in cross-board tests, two ChronoFabric boards fired their packet bursts at exactly the same hardware tick — a measured collective departure spread of 0 ns.

Arrivals hold at the receiving board: in the full transmit-and-receive dataset (10,000 packets, May 2026), arrivals held a trimmed standard deviation of 6.2 nanoseconds in quiet conditions and 6.3 nanoseconds with the processors under stress — inside one 10-nanosecond hardware clock tick. Ungated sending under the same stress degraded to millisecond-scale arrival spread.

A faster cadence, two boards: coordinated scheduling across two boards also held at a 100-microsecond interval — departure spread measured 0.0 nanoseconds, as at 1 ms.

Bursts, delivered inside their window: in burst tests, complete groups of packets were in place ahead of their scheduled instant and every packet departed inside its window — 69 of 69 in a 100 KB group, all inside an 85-microsecond window in the window-enforcement run, and 700 of 700 in a 1 MB group, the larger burst finishing with about 115 microseconds of its 1-millisecond window to spare.

The ring-step pattern: in a two-board ring test — the step pattern a collective operation repeats — scheduled steps held a median and 99th-percentile round trip of exactly 1 millisecond across 999 steps, while unscheduled steps took about 645 microseconds longer per step on average (median round trip about 1.65 ms).

Validated at tighter budgets: the timing mechanism was also validated against the tighter budgets that faster links would impose. On a single KR260 board running its native 1-gigabit connection, the per-packet timing budget was compressed to the cadence equivalents of 100-gigabit (10,000 ns), 400-gigabit (2,500 ns), 800-gigabit (1,250 ns), and 1.6-terabit (620 ns) links — 5,000 packets at each level, in quiet conditions and under full CPU stress — and at every level the schedule held exactly: 100% of packets on target, none stale, no measured spread against the target schedule. This validates the timing mechanism at those budgets; it is not a measurement of data transport at those line rates.

Verified in simulation at the same budgets: the same budgets were also verified a second way — in simulation, against the gate design itself, before any high-rate hardware existed to run them on. In a cocotb regression suite — cocotb is the standard open-source Python framework for verifying chip designs, driving the gate's design description in the GHDL simulator — the gate held its precision at every simulated link-rate budget from 1 gigabit to the 1.6-terabit equivalent: departure spread of zero at all rates, and zero missed gates at the tightest, 620-nanosecond budget. The suite's formal verification report (June 2026) records 45 of 46 tests passing, one skipped, and none failing. That is verification of the design in simulation — not a physical link at 100G, 400G, 800G, or 1.6T — and the full treatment is carried in the companion document, ChronoFabric Measured Results.

A separate equivalence run, on the two-board link: on the two-board 10-gigabit link, the same scheduling ran at a 25-microsecond interval — the timing budget of a 400-gigabit link — for 500 packets with zero misses, holding departure jitter to about 2.4 microseconds at that cadence. That is an equivalent timing budget exercised on a 10-gigabit link, not a 400-gigabit measurement.

ChronoFabric is available now as soft IP — a licensable hardware IP block — targeting Kria FPGA platforms. Embedded chiplets and merchant silicon are future offerings on the staged hardware roadmap — not part of the measurements above.

Staged Hardware Strategy

1

FPGA Evaluation Platforms

ChronoFabric begins on FPGA evaluation platforms — reconfigurable hardware that lets beta customers validate hardware enforcement in their own environments. It is delivered today as soft IP — a licensable hardware IP block — targeting Kria platforms.

2

Production Hardware

From evaluation to production hardware: dedicated devices that carry the fabric into production deployments.

3

Embedded Chiplets & Merchant Silicon

The long-term path: embedded chiplets and merchant silicon for next-generation AI infrastructure, so deterministic orchestration ships inside the platform itself.

Beta & Partnerships

The ChronoFabric beta is open to organizations building or operating large-scale AI infrastructure — teams for whom synchronization waste across accelerators is a real cost, and who want deterministic enforcement evaluated on their own workloads. Beta participation is arranged directly with our team; there is no public download and no form to fill in.

ChronoFabric, like ChronoHive, is proprietary technology of Layer1Labs Silicon Inc., offered for commercial licensing by written agreement.

10 patents pending.

Frequently Asked Questions

What is ChronoFabric?

ChronoFabric is the hardware tier of the Layer1Labs architecture. The same Lingua Franca programs and the same constraint specifications that run in the ChronoHive software engine run on ChronoFabric, with execution enforced on FPGA and hardware endpoints. ChronoFabric is available now as soft IP — a licensable hardware IP block — targeting Kria FPGA platforms. Embedded chiplets and merchant silicon are future offerings on the staged hardware roadmap.

Is ChronoFabric available now?

Yes. ChronoFabric is complete and ready for beta customers. Its admission, sequencing, and determinism guarantees carry over from the software tier unchanged — only the effector binding changes. ChronoFabric is available now as soft IP — a licensable hardware IP block — targeting Kria FPGA platforms. Embedded chiplets and merchant silicon are future offerings on the staged hardware roadmap.

Does ChronoFabric publish benchmark results?

Measured on hardware: in validation runs between two Kria FPGA boards over a 10-gigabit Ethernet link, ChronoFabric delivered more than 21,000 scheduled packets with zero missed departures and zero dropped — a 100% delivery rate, including a continuous 10,000-packet endurance run. The timing mechanism was also validated at the tighter budgets faster links would impose — up to the 620-nanosecond cadence a 1.6-terabit link would require: at every level the schedule held exactly, which validates the timing mechanism at those budgets, not data transport at those line rates. The gate design was also verified in simulation at the same budgets — in a cocotb regression suite driving the gate's design description in the GHDL simulator, the formal report (June 2026) records 45 of 46 tests passing and none failing — verification of the design at those budgets, not a physical link at those rates. The 24-hour run was completed for all three software configurations on a Kria KV260 board: the full hardware profile passed (worst case 22 microseconds), the real-time software configuration passed on its own (33 microseconds), and the stock configuration reached a worst case of 904 microseconds — the gap the real-time configurations close. ChronoFabric is available now as soft IP targeting Kria FPGA platforms; embedded chiplets are a future offering on the staged roadmap.

Who is the ChronoFabric beta for?

The beta is for organizations building or operating large-scale AI infrastructure that want deterministic orchestration enforced in hardware. Participation is arranged directly with our team.

How do we join the beta or discuss licensing?

Email licensing@layer1labs.ai for partnerships and licensing, or info@layer1labs.ai for technical and beta questions. ChronoFabric is offered for commercial licensing by written agreement.

ChronoFabric's status and guarantees, in the technical track → /faq/#technical