🔒
Confidential · Internal & Partner Access

Restricted technical reference

This page contains forward roadmap architecture for base-2 through GF(256) encoding — decimal frequency-bin computing, QUASAR SUPER/THETA, and CV/DV field coding — proposal-grade material kept out of public marketing. Enter the access password to continue.

Confidential · Roadmap Architecture

The path from Base 2 to GF(256)

A forward vision for Spectral Decimal Computing — from today's base-2 dual-rail through native base-10 frequency-bin arithmetic on the Decimal Quantum Frequency Processor (DQFP) — climbing to GF(64) = GF(2⁶), GF(128) = GF(2⁷) on QUASAR SUPER, and CV/DV GF(256) = GF(2⁸) byte-native field coding on QUASAR THETA. Squeezed-light hybrid readout becomes co-primary above GF(64); a nine-layer preservation stack and comb-derived OPC — continuous mesh at scale — on the same Si₃N₄ / TFLN / As₂S₃ platform.

Base-2 → decit GF(2⁶) · GF(2⁷) · GF(2⁸) CV/DV hybrid 9-layer preservation Comb-derived OPC QUASAR ladder
§1 · Roadmap Ladder

Base-2 → Base-10 → GF(64) → GF(256)

Six rungs. One photon. Increasing dimension — not increasing wire count.

Tap or hover any rung to expand. GF(64/128/256) are characteristic-2 finite fields for coding and algebra — distinct from decimal base-10 arithmetic. Rungs R4–R5 are Roadmap / Speculative.

0

Encoding progression

Rungs 0–5 · dimension lift on one spatial mode
R0Today

Base-2 dual-rail qubit

One photon in one of two Si₃N₄ waveguides: |ψ⟩ = α|path A⟩ + β|path B⟩. Hilbert space ℂ², 1 bit/photon. The v1 shipping design: 8-mode Clements mesh, measurement-based fusion, periodic OPC lattice.

d=2 · SU(2) · dual-rail Sagnac OPC demo path

How we do it: Same platform documented on How We Do It. Dual-rail is the seed for GAP04 transcoding at the encoding boundary.

Dual-rail qubit in two waveguide paths
R1Decit

Base-10 frequency-bin "decit"

Transcode path → color (GAP04), then open the comb: |ψ⟩ = Σ_{n=0}^{9} c_n|f_n⟩. Native decimal digit 0…9 without binary decomposition. log₂10 ≈ 3.322 bits/photon. SU(10) gates via EOM + pulse-shaper quantum frequency processor.

D=10 carrier demonstrated externally (Kues 2017) · QLT integration roadmap

How we do it: 50 GHz comb spacing, 10-ring shaper SPU, TFLN EOM FAU, 10-ch WDM demux. OPC at every boundary conjugates all ten colors in one >12 THz pass.

Ten frequency bins of one photon forming a decit
R2Prime power

d=16 qudit (GF(16) engine)

Scale to d=16 = 2⁴ — four bits/photon with cleaner qudit-QEC and mutually-unbiased-basis structure than d=10 (ℤ₁₀ is a ring, not a field). Decimal I/O via transcoder at the application layer (G04).

How we do it: Same QFP stack with 16-ring shaper bank; octet tiling begins here before full 64-line scaling.

Sixteen frequency bins on one waveguide
R3GF(64)

GF(64) = GF(2⁶) field layer

Physical carrier: d=64 frequency-bin qudit on QUASAR NOVA. Logical algebra: GF(2⁶) — XOR addition, log-table multiply, primitive polynomial x⁶+x+1. Unlocks RS/BCH/LDPC, X₆₄/Z₆₄ stabilizers, mature telecom coding IP. Not base-64 decimal.

6 bits/photon · Δf=50 GHz · B_comb=3.15 THz · ~4× OPC margin

How we do it: 64-line locked comb @ Δf=50 GHz, GF(64) FAU (XOR + GFMUL), 4×16 cascaded AWG demux, RS(63,55) encoder, qudit stabilizer rounds via project-then-detect.

Sixty-four frequency bins labeled as GF(2^6) field elements
R4Roadmap

GF(128) = GF(2⁷) — QUASAR SUPER

|ψ⟩ = Σ_{n=0}^{127} c_n|f_n⟩ on a 128-bin carrier. Field layer GF(2⁷) with p(x)=x⁷+x+1 — 7 bits/photon. SUPER sits in QUASAR Region II: the "crisis rung" where one-pass OPC margin tightens to ~2.7× and 16,256 directed phase pairs demand distributed OPC + mandatory squeezing.

Δf=35 GHz · B_comb=4.45 THz · RS(127,119) · [roadmap]

How we do it: 128-tooth comb; 8×16 AWG demux tree; ≥6 dB squeeze per octet; continuous OPC mesh begins here; X₁₂₈/Z₁₂₈ stabilizers with hybrid homodyne+SPAD readout.

R4 GF(128) rung tile
R5Speculative

CV/DV GF(256) = GF(2⁸) — QUASAR THETA

Byte-native symbol field: GF(2⁸) with p(x)=x⁸+x⁴+x³+x²+1 (0x11B) — aligns with RS(255,223), QR, DVB, CCSDS byte algebra. THETA in Region III Q≈1: CV stabilization is co-primary with DV gates; one-pass OPC fails at nominal pitch — continuous mesh + split-band required.

8 bits/photon · Δf=25 GHz · B_comb=6.38 THz · [speculative-framework]

How we do it: 256-line dense comb; 16×16 K×16 demux ladder; ≥10 dB distributed squeeze; global φ₀ calibration plane; RS(255,223) + X₂₅₆/Z₂₅₆ stabilizers; CV-assisted syndrome extraction.

R5 GF(256) rung tile
Six-rung encoding ladder from base-2 to GF(256)Figure 1
Figure 1. The six-rung encoding ladder. Dimension increases on a single spatial mode; comb span grows from 3.15 THz (GF(64)) through 4.45 THz (GF(128)) to 6.38 THz (GF(256)). Gates, detection, and calibration set the ceiling; OPC + squeezing remove the phase floor — continuous mesh beyond GF(64).
§2 · DQFP Instruction Set

Decimal Quantum Frequency Processor

Native decimal arithmetic in SU(10) — not binary decomposition.

ISA

DQFP blocks & opcodes

Register file · FAU · SPU · QFR · FMC
ArchBlocks

Processor architecture

Spectral Register File R₀…R_N — decits on Δf-spaced comb slices. FAU — decimal add/sub/mul/div mod 10 via frequency translation + spectral interference. SPU — 10-ring microring shaper for Z₁₀ and DFT10. QFR — WDM crossbar for multi-decit ops. FMC — SFWM microring memory banks.

How we do it: Lukens QFP model on TFLN-bonded Si₃N₄; FPGA BRAM microprogram tables drive EOM RF and shaper DAC masks.

DQFP five functional blocks on photonic die
Fmt32-bit

Instruction format & decit primitive

| Opcode (8) | SrcA (8) | SrcB (8) | Dst (8) |  → 32 bits
Decit: |ψ⟩ = Σ_{n=0}^{9} c_n |f_n⟩ ,  measurement → digit n ∈ {0…9}

How we do it: T-B07 smoke test: DADD R7,R4,R3 on live 50 GHz comb; herald + project-then-detect validates digit readout.

32-bit DQFP instruction word layout
Arithℤ₁₀

DADD · DSUB · DMUL · DDIV

Decimal addition mod 10: (R_a + R_b) mod 10 via EOM X₁₀^k frequency translation + shaper wrap masks. Worked example: 7 + 4 = 1 (carry = 1) — three-stage MIX → WRAP → X₁₀ spectral trace.

How we do it: FAU east wing on Si₃N₄ bus; microprogram ROM; OPC brackets every arithmetic macro; SDCLA handles multi-digit carry (§3).

Spectral trace of decimal 7+4=1 with carry
ShiftX₁₀

FSHIFT · FCYCLE · PHASE · DFT10

FSHIFT — |f_n⟩ → |f_{n+k}⟩ via TFLN serrodyne at f_RF = Δf. FCYCLE — native X₁₀ increment. PHASE — Z₁₀: |n⟩ → ω^n|n⟩ on shaper only. DFT10 — decimal quantum Fourier transform; Hadamard analog for base-10.

How we do it: SPU 10-ring bank for PHASE; EOM→shaper→EOM cascade for DFT10 (12–32 QFP sandwiches).

FSHIFT serrodyne and Z10 phase mask on ten bins
MultiQFR

C-DADD · SWAP10 · ENTANGLE10

Controlled decimal add, register swap, and maximally entangled decit state (1/√10)Σ|n,n⟩. ENTANGLE10 enables teleport-based COPY (no-cloning respected — G44).

How we do it: QFR WDM crossbar + beamsplitter tiles; same hardware as gate mesh, reprogrammed for multi-register ops.

Maximally entangled two-decit state across frequency bins
ExtAI/QEC

Memory · SYNDROME · DOT10 · MATMUL10

LOAD/STORE to FMC microring banks; SYNDROME/CORRECT/RESET for decit QEC (ℤ₁₀ stabilizers — distinct from GF(64) L7). AI extensions: DOT10, MATMUL10, ATTEND10, FFT10 as spectral macros. Softmax stays classical in v1.

How we do it: FMC L0–L3 memory tiers; QFP U_k before demux for syndrome extraction; AI macros as firmware compositions over F₁₀ + pointwise multiply.

Spectral AI macro pipeline FFT10 DOT10 MATMUL10
§3 · Spectral Carry-Lookahead

SDCLA — decimal arithmetic without ripple delay

Carry propagation as spectral interference, not wire routing.

Generate & propagate

Gᵢ and Pᵢ operators

Gᵢ = 1 iff aᵢ+bᵢ ≥ 10; Pᵢ = 1 iff aᵢ+bᵢ = 9. Carry: C_{i+1} = Gᵢ ∨ (Pᵢ ∧ Cᵢ). All {Gᵢ,Pᵢ} computed in parallel via identical EOM→shaper→EOM GP cells on WDM-separated decit lanes.

How we do it: TFLN beamsplitter GP tiles; shaper masks Θ_G/Θ_P; phase-flag carry encoding to FPGA.

O(log₁₀ N) tree

Spectral carry tree

After parallel GP prediction, radix-10 prefix scan resolves carries in ⌈log₁₀ N⌉ tree levels — vs O(N) ripple. For N=8 decits: ~3 QFP depths vs 8 ripple (~2.7×).

How we do it: FAU east wing combine tiles; associative (G,P)⊗ merge at each tree node; fair-comparison discipline vs binary CLA (G15).

Spectral decimal carry-lookahead treeFigure 2
Figure 2. Spectral decimal carry-lookahead. Each decimal digit occupies an independent frequency channel; carry information propagates through spectral phase relationships rather than physical wiring.
§4 · GF(64) Architecture

What GF(2⁶) unlocks — and its limits

64

GF(64) subsystems

FAU · RS/BCH/LDPC · stabilizers · SU(64) gates
FAUChar-2

GF(64) arithmetic unit

Addition = XOR on 6-bit labels. Multiplication via log/antilog ROM or polynomial convolution mod p(x)=x⁶+x+1. Bins labeled |f_k⟩ ↔ α^k.

How we do it: Reuse DQFP EOM+shaper stack in GF64 mode; electronic shadow on FPGA for validation before full photonic MAC.

GF(64) arithmetic unit XOR and log-multiply
ECCL6

Reed–Solomon · BCH · LDPC

RS(63,55) default for compute registers; BCH(63,45) for FMC pages; LDPC(4096,3584) for bulk/AI tiles. Syndromes are field elements ∈ GF(64) — not decits, not bin indices.

How we do it: Photonic RSENC on G16 FAU; electronic Berlekamp–Massey decode; herald-aware erasure export from L1.

RS(63,55) encode syndrome decode pipeline
QECL7

X₆₄ · Z₆₄ stabilizer codes

63 X-type + 63 Z-type generators per site. Syndrome = QFP rotation U_k → 64-ch project-then-detect → s ∈ GF(64) → CORRECT within ~11.2 ns feed-forward.

How we do it: STAB_LUT BRAM; sparse round scheduling; threshold to-be-modeled (A507 d=64 extension).

GF(64) stabilizer syndrome extraction loop
GatesSU(64)

Universal gate set & AI mapping

{X₆₄, Z₆₄, F₆₄, BS₂} generate SU(64). F₆₄ at 12–32 QFP stages. MATMUL₆₄/ATTEND₆₄: FFT64 → GFMUL → XOR accumulate. Production stays Tier A/B (≤32 depth).

How we do it: Octet-tiled 8×8 shaper banks; offline-compiled F₆₄; OPC every 8–12 stages; softmax classical at host.

SU(64) QFP cascade depth timeline
Handoff · where GF(64) stops being comfortable

GF(64) on NOVA is the last rung where one-pass comb-derived OPC sits in a comfortable ~4× margin band and 64-ch SPAD readout suffices. Above d=64 the STAR-PHASER linear paradigm breaks: QUASAR takes over — tighter Δf, mandatory squeeze, hybrid homodyne syndromes, continuous OPC mesh, and calibration that scales as N(N−1) directed phase pairs (16,256 @ d=128; 65,280 @ d=256). See QUASAR framework Law I–II and §10 below. Roadmap

§5 · Squeezed Light & DV/CV Hybrid

Frequency labels tensor with squeezed qumodes

Squeezing does not change carrier frequency — it tightens the quantum noise floor.

Each comb tooth is a harmonic oscillator: |n⟩_freq ⊗ |ξ_n⟩_CV. DV gates act on the color index; squeezing + OPC act on quadratures within each color. OPC reverses coherent drift (φ→−φ); squeezing attacks quantum uncertainty (sub-SQL syndromes, LIGO analogy).

Practical near-term use

Phase-squeezed frequency-bin states improve spectral gate fidelity, QEC syndrome measurement, and readout — not an additional radix dimension. Speculative (f_n, r_n) two-DOF encoding is a documented hedge only (G30).

On-chip stack

How we do it

  • χ⁽³⁾ Si₃N₄ microring squeeze sources (4–7 dB inferred)
  • Ge p-i-n balanced homodyne @ 15.3 GHz, pump-split LO
  • Dual fork: SPAD (population) + homodyne (phase syndromes)
  • GKP/cat/binomial bosonic codes on selected bins (L5)
  • Scale: 64 squeeze channels on GF(64) roadmap
DV frequency-bin and CV squeezed-qumode hybridFigure 3
Figure 3. DV frequency-bin + CV squeezed-qumode hybrid. One bus, one pump tree; quadrature readout and photon counting share the same demux.
§5.1 · CV scaling ladder

Squeezing requirement vs dimension

σ² = ½e^(−2r). Optional at GF(64); mandatory from GF(128) upward.

CV

Squeeze staircase

3 dB helpful → 6 dB mandatory → 10 dB distributed
d≤64Helpful

3–6 dB squeeze at GF(64)

χ⁽³⁾ Si₃N₄ microring sources (4–7 dB inferred) tighten phase syndromes and gate fidelity. OPC still carries primary phase hygiene; squeeze is a performance multiplier, not yet mandatory. Roadmap

How we do it: Sparse squeeze allocation on compute octets; Ge homodyne @ 15.3 GHz for Z-syndrome taps.

Optional squeeze GF(64)
d=128NS-05

≥6 dB mandatory per octet

At SUPER, 35 GHz pitch and 16,256 directed phase pairs collapse OPC margin to ~2.7× WARN. OPC + squeeze mandatory — external precedent: 18 dB TFLN, 7.5 dB TF-PPLN microresonator (2025–2026). Roadmap

How we do it: One squeeze source per octet (16 octets); pump-null schedule; squeeze verifies before grid_valid.

Mandatory squeeze GF(128)
d=256Distributed

≥10 dB distributed at GF(256)

THETA Region III: sub-SQL syndromes and GKP-qudit bridges require distributed ≥10 dB across 32 octets with global φ₀ reference. CV sector is co-primary with DV gates — not a sidebar. Speculative

How we do it: Stitched squeeze mesh; continuous OPC boundary layer; tensor |f_n⟩⊗S(r) on every compute bin.

Distributed squeeze GF(256)
Squeezing-vs-d staircaseFigure 3a
Figure 3a. CV sector scaling ladder. Squeezing tightens the quantum noise floor within each frequency bin; requirement escalates from optional (GF(64)) through mandatory (GF(128)) to distributed (GF(256)).
§5.2 · Hybrid readout ladder

Project-then-detect at 64 → 128 → 256 channels

L1–3

Readout tiers

SPAD tree → hybrid homodyne → CV-assisted φ₀
L1GF(64)

64-ch SPAD project-then-detect

Baseline: QFP rotation U_k → 4×16 AWG demux → warm SPAD array. Population readout for X-syndromes; homodyne on selected Z generators. Designed/target

L1 64-ch SPAD tree
L2GF(128)

128-ch hybrid homodyne + SPAD

64 SPAD + 64 Ge homodyne pairs in dual-path syndromes — DV digit + CV Z-syndrome on same demux fork. 8×16 routing tree. Roadmap

L2 hybrid 128-ch
L3GF(256)

256-ch CV-assisted + global φ₀

Full tensor |f_n⟩⊗S(r) readout with global phase reference plane; compiler classifies each stabilizer row as SPAD / HOMODYNE / DUAL. 16×16 demux ladder. Speculative

L3 CV-assisted 256-ch
Hybrid readout ladderFigure 3b
Figure 3b. DV/CV hybrid readout ladder. Readout complexity scales with d: population counting alone cannot carry Z-syndromes at GF(128)+; homodyne arrays and a global φ₀ plane become architectural, not optional.
§6 · Nine-Layer Preservation

How we preserve quantum data on the chip

Each layer removes a different error class before accumulation.

Tap any layer. L8 (topological spectral) is speculative; L9 orchestrates refresh across L1–L8.

L1–9

Preservation hierarchy

Photon → field symbol → logical qudit → encoded state
L1Photon

Ultra-low-loss Si₃N₄

α < 0.01 dB/cm target; As₂S₃ quarantined to OPC spirals only. Survival S = 10^(−L/10). Herald erasures feed RS decoder.

How we do it: Damascene SiN backbone; Euler bends; split-fab BEOL overlay only at OPC nodes.

Ultra-low-loss silicon nitride routing
L2Bins

Spectral orthogonality

Comb locking: SFWM birth coherence + per-tooth heater grid. Guard bands between compute bins, ancilla, pumps. Leakage budget F_orth ≥ 0.999.

How we do it: L0 global → L1 octet → L2 tooth → L3 TED firmware loops; grid_valid gate.

Locked frequency comb with guard bands
L3OPC

Relative phase + OPC

Broadband FWM: φ_n → −φ_n for all d bins in one pass. Periodic lattice every ~10 gates; boundary OPC at GAP04 transcoder. Error bounded by interval M, not total depth.

d≥128 stress: One-pass margin WARN @ GF(128); FAIL @ GF(256) — continuous boundary mesh replaces monolithic spiral (§7.1).

How we do it: Hybrid As₂S₃/SiN recirculating spirals; comb-derived pumps (§7).

OPC phase conjugation across all frequency bins
L4DFS

Decoherence-free subspaces

Encode in relative spectral coordinates (Δn, φ_nm) not absolute lab frequency. Global comb slip handled by CAL_GLOBAL_SLIP retune — not syndrome.

Relative spectral encoding invariant to common-mode drift
L5CV

Bosonic encoding

|n⟩_freq ⊗ |ψ_L⟩_CV — binomial (v1 bridge), cat (Z-biased channel), GKP-qudit (GF(64) roadmap). Sparse allocation policy TBD.

d≥128 stress: Squeeze mandatory per octet (NS-05); GKP-qudit bridges become load-bearing at GF(256), not sparse hedge.

Bosonic codewords on frequency-bin modes
L6ECC

GF(q) classical codes

RS/BCH/LDPC over GF(64) → RS(127,119) @ GF(128) → RS(255,223) @ GF(256). Photonic parity generation; electronic Berlekamp–Massey decode. Syndromes are field elements — firewall from decit ℤ₁₀ QEC.

d≥128 stress: Syndrome width doubles each rung; photonic encode stays on-chip, BM decode stays electronic.

Classical error correction over GF(64)
L7QEC

Qudit stabilizer codes

X_d/Z_d rounds via QFP project-then-detect: 126 → 254 → 510 generators per site as d climbs. OPC biases Z-errors; digital qudit QEC handles residual. Not a replacement for OPC.

d≥128 stress: Hybrid readout (§5.2) required; stabilizer rounds interleave with L9 refresh slots.

d=64 qudit stabilizer quantum error correction
L8Speculative

Topological spectral ring

SSH dimer couplers, Lieb zero-modes, NHT-OPC ring closure — research option only. Kill criterion T-G39-01. Not v1, not gating.

Speculative topological protection on frequency ring
L9Refresh

Dynamic spectral refresh

Five synchronized clocks: gate OPC, FMC DRAM, RS re-encode, bosonic rounds, stabilizer slots. G40 firmware arbiter — coherent DRAM refresh analogue.

d≥128 stress: Refresh interleaving becomes tighter; continuous OPC mesh shares the same arbiter as squeeze pump nulling.

How we do it: refresh_state BRAM; compiler RefreshPass; priority arbiter on photonic bus.

Dynamic spectral refresh orchestration
Nine-layer stress d=128Figure 2a
Figure 2a. Nine-layer stress @ d≥128. Layers L3 (OPC), L5 (CV), L6 (ECC), L7 (stabilizer), and L9 (refresh) strain first as directed phase-pair count grows N(N−1).
§7 · FWM Pump & Comb-Derived OPC

What the pump does — and how OPC works with it

Phase conjugation

ωᵢ = 2ωₚ − ωₛ · φ → −φ

E_s = A_s e^{iφ_s}  →  E_i ∝ A_s e^{−iφ_s}
Pump supplies energy + phase reference — not quantum information.
OPC is unitary Bogoliubov evolution — not cloning (no-cloning theorem).

How we do it: Hybrid As₂S₃/SiN spiral; pump in 7.4 THz Raman-null window; ≥110 dB pump rejection; η scales ~P_pump².

Comb-derived OPC

Teeth as pumps — one coherent source

Reserve interior comb teeth as degenerate FWM pumps for neighbors. Logic channels, memory, and OPC pumps share SFWM birth coherence — no independent laser sync. 8 pump sites at d=10; 62 at d=64. Spectral re-mapping returns idler to GF(64) α^k grid.

How we do it: opc_pump_mask firmware; octet-stagger if inter-pump crosstalk > −25 dB; f₆₃ guard (α⁶³=α⁰).

Comb-derived OPC mesh on 64-tooth gridFigure 4
Figure 4. Comb-derived OPC. The microcomb that births computational bins also supplies the pumps that conjugate them — common-mode phase stability by construction.
§7.1 · OPC margin ladder

One-pass comfort → WARN → FAIL

Law I: m_margin = B_OPC / B_comb. Continuous mesh when one-pass fails.

PASS · GF(64)

~4× comfortable

Δf=50 GHz · B_comb=3.15 THz. Monolithic one-pass OPC on hybrid As₂S₃/SiN spiral covers all 64 bins. Last rung where periodic lattice alone suffices.

WARN · GF(128)

~2.7× tight

Δf=35 GHz · B_comb=4.45 THz. SUPER needs distributed OPC cells every Λ=2–5 mm + mandatory squeeze. Comb-derived pumps staggered per octet.

FAIL · GF(256)

One-pass fails

At Δf=50 GHz, B_comb=12.75 THz > 12 THz claim → 0.94× FAIL. THETA uses Δf=25 GHz (6.38 THz) + continuous boundary mesh + split-band A∥B per QUASAR NS-14.

How we do it: GALAXY sparse OPC cells → NOVA periodic lattice → SUPER distributed mesh → THETA continuous boundary layer (Λ cells, comb-derived pumps, split-band). Si₃N₄ FWM bands ~1–6 THz demonstrated; QLT hybrid >12 THz remains to-be-tested (T9).

OPC margin waterfallFigure 4a
Figure 4a. OPC scaling ladder. Margin collapses with comb span; GF(256) requires continuous mesh architecture — not a wider one-pass spiral.
Continuous OPC meshFigure 4b
Figure 4b. Continuous OPC mesh. Boundary-layer cells conjugate phase locally; split-band architecture covers LOTUS-scale spans without a single monolithic pass.
§8 · How We Build It

GF(64) tile on the QLT process flow

Same materials stack as the decit. Scaling d=10 → d=64 is a manufacturing problem solved by three physical levers — not a new platform.

Comb + QFP

64-line microcomb & octet QFP

R≈480 µm SFWM ring or EO-comb @ Δf=50 GHz (3.15 THz span). 8× octet-tiled TFLN EOM → 8-ring shaper cascades. Same B07 materials stack as decit — scale channels, not platform.

Uniformity

Wafer scatter + thermal trim

Accept 5–15% resonance scatter; lock every tooth with hierarchical heaters (L0 global → L1 octet → L2 tooth → L3 TED). T-G47 signoff matrix.

Readout

64-ch demux + dual detection

Cascaded 8×8 AWG; warm SPAD array for digit read; Ge homodyne for phase syndromes. Project-then-detect for any basis. F_read ≈ 0.998 cascaded target.

GF(64) photonic tile floorplanFigure 5
Figure 5. GF(64) photonic tile floorplan. Three co-integrated blocks on one Si₃N₄ die: locked 64-tooth comb, octet-tiled QFP cascade, cascaded 64-ch WDM — with As₂S₃ OPC after the FAU boundary.
§8.1 · The physical build

Three levers — d=10 through d=256

Tap any lever. None of this is a new materials stack — it is more of the same device, cascaded routing, and active calibration. Columns show scaling at d=64 (NOVA), d=128 (SUPER), and d=256 (THETA). [roadmap / speculative-framework — Rungs R3–R5, not shipped]

Leverd=64 NOVAd=128 SUPERd=256 THETA
L1 Modulators12–32 EOM sandwiches · ~32 RF tones32–48 · 16 octets · 48 RF tones48–64+ · 32 octets · 64 tones
L2 Routing4×16 AWG @ 50 GHz8×16 @ 35 GHz16×16 @ 25 GHz DWDM
L3 CalibrationN(N−1)=4,032 pairs · ChiL/octet16,256 pairs · ChiL++65,280 pairs · global φ₀ plane
×3

Scaling levers

More modulators · larger routing · better calibration
L1Modulators

More modulators — deeper QFP cascade

GF(64) is the same TFLN modulator, used more times. Named Tier-A/B gates compile to 12–32 EOM→shaper→EOM sandwiches (vs 3–5 for the decit), driven by a harmonic RF bus of up to ~32 tones (f_m = m·Δf) on one segmented traveling-wave modulator. At d=128/256 depth grows to 32–48 → 48–64+ sandwiches with 16 → 32 octet shaper tiling. Field precedent: 256-channel TFLN arrays and 600+ phase modulators on 200 mm wafer [demonstrated].

How we do it: Segmented TFLN bond (Vπ·L ≈ 2.5 V·cm); octet shaper banks; As₂S₃ OPC every ~10 stages; RF tables frozen in FPGA BRAM. Roadmap @ d≥128.

Deeper EOM-shaper-EOM cascade with harmonic RF bus
L2Routing

Larger routing networks — K×16 ladder

Scale by multiplying 16-ch AWG blocks, not one giant demux: 4×16 (NOVA @ 50 GHz) → 8×16 (SUPER @ 35 GHz) → 16×16 (THETA @ 25 GHz DWDM). Each level: octet add-drop → AWG fan-out → Vernier trim, targeting ε ≤ −30 dB cascaded crosstalk. 3D interlayer crossings cut per-crossing loss ~100× [demonstrated].

How we do it: EO reconfigurable TFLN Vernier filters for ns trim; herald switch tree replicated per octet; AWG slices co-calibrated with shaper octets. Roadmap @ d≥128.

Cascaded K×16 WDM routing ladder
L3Calibration

Better calibration — N(N−1) pairs → φ₀

Pairwise phase budget scales N(N−1): 4,032 @ d=64 → 16,256 @ d=128 → 65,280 @ d=256 (directed convention). Thermal matrix grows 64² → 128² → 256². Mode escalates: ChiL/octet → ChiL++/cross-octet → global φ₀ plane at THETA.

How we do it: Chip-in-the-loop self-calibration (O(kN)→O(1) in ~10 iterations) [demonstrated]; grid_valid gate blocks compute until |Δf_err| < 0.5 GHz; WDM drops co-trimmed with shaper octets. Speculative @ φ₀ plane.

ChiL to global phi-zero calibration architecture
Deeper EOM-shaper-EOM cascadeFigure 6
Figure 6. More modulators. The d=64 QFP is the decit cascade made deeper — 12–32 EOM→shaper→EOM sandwiches on a shared harmonic RF bus, OPC-bounded to 6–10 dB effective loss.
64-channel WDM routing treeFigure 7
Figure 7. Larger routing networks. Octet add-drop → 4×16 SiN AWG → per-bin Vernier sorts 64 bins at −30 dB cascaded crosstalk; 3D layers cut crossing loss ~100×.
Three-loop hierarchical comb-lock calibrationFigure 8
Figure 8. Better calibration. Global TEC → octet → per-tooth heaters lock 64 teeth (|Δf_err| < 0.5 GHz); chip-in-the-loop self-calibration converges in ~10 iterations.
§10 · Beyond GF(64)

The CV/DV road to GF(256)

Synthesis: why the ladder continues, what changes at each crisis rung, and where to read deeper.

Canonical Δf table (S84)
RungChipΔfB_combRegionOPC margin
R3 GF(64)NOVA50 GHz3.15 THzRegion I~4× PASS
R4 GF(128)SUPER35 GHz4.45 THzRegion II~2.7× WARN
R5 GF(256)THETA25 GHz6.38 THzRegion III Q≈1mesh required

Brand boundary at d=64: TETRIS → NOVA; STAR-PHASER linear paradigm → QUASAR field-computing architecture. Q-metric tracks coherence budget vs dimension — see QUASAR framework §4.

§10

Extension topics

Encoding · CV · OPC · QEC · routing · applications
10.1GF(128)

R4 — GF(2⁷) on 128 bins

7 bits/photon; p(x)=x⁷+x+1; RS(127,119); 16,256 directed phase pairs. Crisis rung where squeeze + distributed OPC become mandatory. SUPER chip page · Roadmap

GF(128) encoding extension
10.2GF(256)

R5 — byte-native GF(2⁸)

8 bits/photon; 0x11B polynomial; RS(255,223) aligns with QR/DVB/CCSDS byte algebra. CV stabilization co-primary. THETA chip page · Speculative-framework

GF(256) byte field
10.9Apps

GF(256) coding payoff

Hybrid classical–quantum byte pipeline: host RS → photonic RSENC → herald → electronic BM decode. RS(255,223) streaming demo concept (~111k sym/s target). Investor framing: GF(256) is where telecom byte IP meets photonic qudit carriers. Roadmap

GF(256) application stack
Related

From roadmap to today's platform

The v1 machine is base-2 dual-rail. See the full photon journey, manufacturing path, and patent portfolio.