Skip to content
LWLG
~13 min read · 3,086 words · confidence 59%

Silicon photonics supply chain — L1–L6 stack with LWLG positioning

Purpose: A standalone L1–L6 reference of the silicon-photonics supply chain, naming the entities at each layer and identifying where Lightwave Logic sits, who LWLG’s customers’ customers are, where competitors sit, and where the choke points are 2026-2028. Designed to complement kb/03_ecosystem/competitors.md (platform-vs-platform) and kb/03_ecosystem/customer_roadmaps.md (per-candidate roadmap).

As-of date: 2026-04-26.

Stack convention: L1 = raw materials and chemistry; L6 = end-customer / hyperscaler deploying the optics in production. LWLG’s primary value-creation seat is L1 (Perkinamine polymer chemistry) with PDK reach into L2 (foundry). Revenue from a single Stage-4 customer flows L1 → L2 → (L3) → L4 → L5 → L6.

L1–L6 supply chain — where LWLG sits

layered stack · arrows = upstream value flow
L6
layer
Hyperscalers / Telcos
GoogleMicrosoftMetaAmazonByteDanceOracle CloudAT&TVerizonDeutsche TelekomNTT
↑ buys modules + integrated systems
L5
layer
Module / OEM
EoptolinkInnolightCoherentLumentumCiscoCienaAristaMarvell (POH via Polariton)
↑ buys + packages PIC + DSP + laser
L4
layer
Packaging / Assembly
JabilFoxconnSenkoUS ConecTFCAmkorASETE ConnectivityPHIXAlter Tech
↑ singulation, fiber attach, fanout
L3
layer
DSP / Driver electronics
BroadcomMarvell (Inphi)Cisco/AcaciaMaxLinearCredoSemtechMACOMAlphawave Semi
↑ SerDes + DSP signal in
L2
layer
Wafer fab / SiPh foundry
LWLG PDK reach
Tower (PH18)GlobalFoundries (Fotonix + AMF)SilTerra (Luceda)TSMC (COUPE)UMC+1 Tier-1 unnamedIHP-MicroelectronicsAIM PhotonicsSmart Photonicsimec (R&D)Fraunhofer HHI
↑ spin-coats polymer + integrates as BEOL
L1
layer
Materials / chemistry
★ LWLG core
★ Lightwave Logic (Perkinamine)NLM Photonics (Selerion)HamamatsuTOKIdemitsuSumitomoBASFDuPontDowFerroAlmega (BTO)NanoLN (TFLN substrate)
↓ Revenue flow direction
A single Stage-4 customer at L5/L6 drives polymer demand all the way down to L1. LWLG earns at the source.
↑ Material flow direction
Perkinamine ships from L1 (Englewood CO) to L2 foundries as a BEOL deposition step, then through L3-L5 into L6 deployment.
Choke points / strategic moats
L1 Reliability data LWLG 2,000h Telcordia 85/85 length-swept (Apr 2026) — single-strongest L1 differentiator vs NLM (no equivalent)
L1 Patent stack ~22 distinct inventions / 35 US grants (validated) · NLM filed thermoset patent Mar 2025 — interlocking IP sets up duopoly
L2 Foundry capacity 4 PDK partners × multi-quarter integration cycles — SiPh foundry capacity is constrained 2026-2028
L2 PDK lock-in Re-tape-out at different foundry costs $1-3M and 6-12 months — LWLG's structural moat
L3 CPO architecture Marvell + Polariton vertical-integrated POH stack now Marvell-owned · LWLG = polymer feed
L5 400G/lane window TFLN currently dominant (HyperLight + Lumera) · LWLG window opens at 800G/lane (3.2T) ~2027-2028

LWLG’s pricing power and recurring-revenue thesis lives at L1; the strategic moat lives in the L1↔L2 PDK-handoff (the Perkinamine has to be qualified inside foundry process flows). Stage-3 customers can sit at L3, L5, or L6 — material specification authority lies wherever the next-generation transceiver is being architected.

L1 — Materials chemistry

The active medium. This is where LWLG creates its IP: proprietary EO polymer chromophores plus the host-glass and processing chemistry that make them survive 85 °C / 85 % RH for 2,000+ hours.

EntityMaterial / productPositionLWLG relationship
★ Lightwave Logic (LWLG)Perkinamine (guest-host EO polymer; Series 3 named in 256 GBd plasmonic ring paper); 4th-gen ALD encapsulation (OTR 1.4×10⁻⁶ g/m²/day)Direct — IP holdern/a
NLM PhotonicsSelerion family (BHX, HTX, JRD1) — crosslinkable thermoset EO polymerDirect competitorBoth academic-rooted (LWLG via Univ. Delaware; NLM via UW/Larry Dalton)
Tokyo Ohka Kogyo (TOK)Photoresist + EO polymer industrial-supply partnerNLM-alignedLWLG ecosystem-adjacent
Hamamatsu PhotonicsIndustrial-supply partnerNLM-alignedLWLG ecosystem-adjacent
Idemitsu KosanSpecialty-polymer supplierNLM-alignedn/a
NanoLN (China)TFLN wafer supplier (substrate-level material) — competes via TFLN, not polymerCompeting platformSubstrate-level competition
Almega / Fraunhofer / IBM Zurich (BTO targets)Barium-titanate filmsCompeting platform (TRL 4-5)Distant
DuPont, Dow, BASF, Sumitomo Chemical, FerroAdjacent specialty-chem supply (ALD precursors, CMP slurries, photoresist additives)Foundry-side suppliersLWLG’s K-POD process likely integrates adjacent-supplier chemistry

Choke point at L1:

  • Reliability data trail. LWLG’s 1,000h Telcordia 85/85 (Jul 2025) → 2,000h length-swept (Apr 2026) is the single-strongest L1 differentiator. NLM has not published equivalent qualification data.
  • Chromophore patent stack. LWLG portfolio is 22 distinct inventions / 35 US grants (primary-source validated Apr 2026); aggregator counts (Minesoft 47/142/184, GreyB 78) reflect jurisdictional counterparts of the same inventions. Continuation chains run 2038-2042. NLM filed thermoset-EO-polymer patent in March 2025. Inter-locking IP risk could shape future licensing.
  • Rare-earth supply. LWLG’s Oct 15 2025 PR explicitly affirmed Perkinamine is fully rare-earth-free with all production in Denver/Englewood CO — geopolitical advantage vs. China-rare-earth-dependent platforms.

2026-2028 outlook for L1:

  • 2026: First production-scale polymer wafer-runs through GF/Tower/SilTerra PDKs (already in tapeout).
  • 2027: Stage 3 → Stage 4 conversion is the trigger event. Cost-target validation.
  • 2028: Polymer + TFLN duopoly probable for 1.6T+ at-pluggable-density and CPO architectures.

L2 — Wafer fab / silicon-photonics foundry

Where the modulator gets built. LWLG’s strategic seat at L2 is its PDK presence in three independent foundries (GF, Tower, SilTerra) with a fourth in mention. The PDK is the forcing function that converts a Stage-2 customer into a Stage-3 customer — once a customer designs around an LWLG-PDK-resident polymer modulator, the switching cost is high.

EntityProcess / platformLWLG statusNotes
GlobalFoundries (Fotonix + AMF Singapore)Fotonix 45CLO + 9WG (300mm); AMF SiPh (200mm; acquired by GF Nov 17 2025)✓ GDSFactory PDK live Mar 16 2026; live PDK demo at OFC 2026LWLG’s flagship 300mm foundry; AMF acquisition consolidated capacity
Tower SemiconductorPH18 SiPh process (200mm)✓ Development agreement Mar 11 2026Compact low-power 200G/400G architectures; Tower-Coherent demo’d 400 Gbps/lane at OFC 2026
SilTerra (Malaysia)SiPh (200mm) via Luceda PDK✓ Initial wafer tapeout completed early 2026; Mar 3 2026 PRFirst Asian-foundry LWLG PDK
TSMCCOUPE silicon-photonics platform⚠ Speculation: LeMaitre referenced “2 unnamed foundries” per IH speculation; not confirmedSpeculation only; TSMC dominates NVIDIA’s CPO partnership
UMCSiPh foundryNot currently LWLG-affiliatedHyperLight’s foundry partner (TFLN) — competing platform anchor
IHP-Microelectronics (Frankfurt-Oder)EU SiPh foundryNot currently LWLG-affiliatedOpen option for European stack
AIM PhotonicsUS R&D + low-volume foundry (Rochester NY)Not currently LWLG-affiliatedGovernment-backed R&D; possible small-run validation
Smart Photonics (Eindhoven)InP foundryNot relevant to LWLG (different platform)InP-only
imec (Leuven)iSiPP200 SiPh R&D platform✓ Polariton has Plasmonic POH on imec iSiPP200 (Dec 2025)LWLG materials feed Polariton; indirectly LWLG-relevant
Fraunhofer HHI (Berlin)Polymer + InP researchAdjacentHistorical EO polymer research collaborator

Choke point at L2:

  • Foundry capacity for SiPh wafers. GF Fotonix and Tower PH18 are commercially running but at limited capacity vs hyperscaler demand 2026-2028.
  • PDK-customer onboarding speed. Per LeMaitre at ECOC 2025, “Several foundries waiting to be onboarded but LWLG doesn’t have enough engineering resources” — engineering staffing is the gating constraint, not foundry capacity.
  • TSMC COUPE. If TSMC adopts polymer (vs TFLN or BTO), the supply curve changes overnight; this is the single-biggest L2 watch item.

2026-2028 outlook for L2:

  • 2026: First polymer-PDK MPW shuttles run on GF + Tower + SilTerra; LWLG validates production yield.
  • 2027: One foundry transitions LWLG PDK from MPW shuttle into volume tape-out cadence (= Stage 4 trigger).
  • 2028: TSMC COUPE adoption decision; if polymer wins, moat compounds rapidly.

L3 — DSP / driver electronics

The DSP/driver pairs with the optical modulator. L3 is where Marvell, Broadcom, Cisco, MACOM compete. L3 doesn’t directly touch LWLG polymer chemistry, but L3 vendors are the architecture authorities for transceiver-system specs, so a Stage-3 LWLG customer at L3 (e.g., Marvell) is the highest-value relationship.

EntityProduct lineLWLG linkNotes
★ Marvell (NASDAQ: MRVL)Inphi PAM4 DSPs · coherent DSPs (800G ZR/ZR+, 1.6T ZR in dev) · custom AI silicon · CPO development · Polariton POH (acquired Apr 22 2026)✓ Strategic via Polariton acquisition; LeMaitre/Chowdhury/Hailemariam liked KPMG announcement on LinkedInInphi $10B acq 2021; received NVDA optical-supply-chain investment 2025-2026
★ Broadcom (NASDAQ: AVGO)Taurus BCM83640 — 3nm 400G/lane optical DSP launched Mar 11 2026 for 1.6T modules; lays foundation for 3.2T optical transceivers and 204.8T switch platforms◐ Indirect — Broadcom DSP could pair with LWLG-materials-based optical front-endMass production late 2026; >100M units 1.6T+3.2T over 5 years
Cisco/AcaciaAcacia 3nm Kibo 1.6T PAM4 DSP; 200G/lane optical engines for AI workloads; coherent pluggable optics for routed optical networking◐ Low — Cisco’s Acuity Photonics path is in-houseAcacia acquired 2021 ($2.6B)
MaxLinearOptical PAM4 DSPs (mid-tier)IndirectLower-volume play
Credo SemiconductorSerDes + DSP IPIndirectHyperscaler custom-DSP supplier
SemtechPAM4 DSPs and CDRIndirectMid-tier
MACOMInP modulator drivers + EML driversIndirectDriver-electronics for InP and SiPh
Alphawave SemiHigh-speed SerDes IPIndirectCustom-silicon enabler

Choke point at L3:

  • DSP-to-modulator interface specs. LWLG’s <1V drive at 100-200 GBaud is below most DSP output swing — actually a positive for power efficiency. Marvell + Broadcom drivers can run at lower voltages, saving power.
  • Marvell vs Broadcom competitive dynamic. Both target 400G/lane 3.2T modules; whichever DSP wins shapes the modulator-material decision indirectly.

2026-2028 outlook for L3:

  • 2026: 1.6T DSP shipping volume; first 400G/lane DSP samples (Taurus).
  • 2027: 3.2T modules in production; first CPO-integrated DSPs.
  • 2028: NVIDIA Feynman NVL1152 all-CPO wave drives full-stack integration.

L4 — Packaging / assembly

Hermetic seal, fiber attach, singulation. L4 is where Jabil, Foxconn, Senko, US Conec, Amkor, TE Connectivity, ASE, TFC operate. LWLG’s 4th-gen ALD encapsulation and gold-box packaging interact with L4 design rules.

EntityPositionLWLG linkNotes
JabilMajor contract packager (TFLN HyperLight singulation/package); SiPh module assembly◐Wildt at OFC 2026: “BTO and polymers not as mature” as TFLN — bear datapoint, though KCCO7913 noted Jabil packaging maturation for TFLN should enable polymer faster
FoxconnHyperscaler-scale module assembly◐Transceiver-volume packager for NVIDIA orders
Senko Advanced ComponentsMPO connectors, fiber-attach hardwareIndirectStandard L4 bill-of-materials
US ConecMPO/MTP optical connectors; high-density fiber arraysIndirectStandard L4 bill-of-materials
TFC (Taiwan Fiber Communication)Optical-module assembler — assembled HyperLight 1.6T-DR8 reference transceiver at OFC 2026Competing-platform packagerDemonstrates TFC capability for TFLN; could extend to polymer modules
AmkorOSAT (Outsourced Assembly + Test); CPO packaging◐Co-packaged-optics assembler; speaker at ISIG Taiwan 2026
ASE Group (Advanced Semiconductor Engineering)OSAT for advanced packaging◐TSMC-aligned OSAT
TE ConnectivityOptical-connector + interposerIndirectIntegration components

Choke point at L4:

  • CPO packaging maturity. Co-packaged optics requires hermetic die-to-die alignment at <1 µm precision; this is where Amkor + ASE + Foxconn are scaling.
  • Hermetic packaging for EO polymer. LWLG’s 4th-gen ALD encapsulation reduces reliance on hermetic gold-box packaging — a direct positive for L4 cost.

2026-2028 outlook for L4:

  • 2026: Pluggable 1.6T + early 3.2T volume.
  • 2027: First hyperscaler-scale CPO deployments (Marvell, Broadcom, Cisco at chip-package level).
  • 2028: NVIDIA Feynman NVL1152 = full-CPO architecture; L4 packaging volume at unprecedented scale.

L5 — Module OEM / transceiver vendor

The pluggable / module brand. L5 is where Eoptolink, Innolight, Coherent, Ciena, Lumentum, Cisco, and Arista compete. LWLG’s polymer enables L5 vendors to hit the lower-power-per-bit / lower-latency target that AI factories require.

EntityPosition2026 capabilitiesLWLG link
EoptolinkTop-2 transceiver volume (with Innolight); demo’d 1.6T DR4 IMDD 400G/lambda + 6.4T NPO + 12.8T XPO at OFC 2026 (booth #1211)1.6T pluggable in volume 2026; 3.2T sampling 2027◐ Indirect — uses Marvell/Broadcom DSPs that pair with optical front-ends; HyperLight TFLN integrator currently
Innolight TechnologyTop-2 transceiver volume (with Eoptolink); dominates 800G SFP modules — 60% combined supply with Eoptolink for NVIDIA orders1.6T volume 2026◐ Indirect
Coherent Corp (NASDAQ: COHR)Vertically-integrated transceiver + optical-component vendor; Demonstrated multiple 1.6T transceivers at OFC 2026 with SiPh PIC, InP CW laser, 200G InP EML, 200G GaAs VCSEL; Demonstrated 3.2T (400G/lane PAM4) with 400G Differential EML + SiPh PIC + 400G pure silicon PN-junction MZM; XPO MSA founding member for 12.8Tbps liquid-cooled optics1.6T multi-flavor 2026; 3.2T 2027; 12.8T XPO 2027-2028◐ Stage-3 candidate (medium); OFC 2026 plenary by CTO Eng on bandwidth-per-watt
Ciena (NYSE: CIEN)WaveLogic 6E coherent (1.6T); WaveLogic 6 Nano with “Coherent-Lite” 1.6T 2-20km DCI; 1.2T+ Ottawa-lab roadmap; CDM today is InP, evaluating TFLNWaveLogic 6 Nano 1.6T avail 2026; 3.2T target 2027-2028◐ Stage-3 candidate (medium); CDM second-source need explicit
Lumentum (NASDAQ: LITE)NVDA $2B strategic investment Mar 2026 for 200G/lane EML capacity; 1.6T DR4 OSFP modules begin full-scale shipping summer 2026; mgmt guidance $2B/quarter run-rate by late 2027; 2028 = CPO + optical-engine pivot1.6T volume H2 2026; 3.2T 2027; CPO engines 2028+◐ Stage-3 candidate (medium-low); NVDA-aligned
CiscoAcquired Acacia (2021) + Acuity Photonics; coherent + IMDD pluggables1.6T module via Acacia DSP◐ Stage-3 candidate (low); in-house photonics path
Arista NetworksSwitch + transceiver buyer (not module vendor itself); top destination for 800G/1.6T modulesCustomer of Eoptolink/Innolight/Coherentn/a
MarvellAcquired Polariton Apr 22 2026 — first time module-vendor seat acquired; previously DSP-only L3POH 2027+ via Polariton✓ Strategic via Polariton acquisition

Choke point at L5:

  • NVIDIA orders. Innolight + Eoptolink dominate 60% of 800G NVIDIA orders; the same duopoly likely scales with 1.6T and 3.2T.
  • CPO transition. L5 vendors are pivoting from pluggable modules to CPO optical engines (Lumentum’s 2028 plan); this is where LWLG-class polymer’s compactness gives a structural edge.

2026-2028 outlook for L5:

  • 2026: Pluggable 1.6T volume; first 3.2T samples.
  • 2027: 3.2T volume; coherent-lite DCI volume; CPO engineering samples.
  • 2028: CPO production starts at hyperscaler; pluggable + CPO co-exist.

L6 — Hyperscaler / end-customer

Where the optics actually deploy. L6 is the demand engine. NVIDIA, Google, Microsoft, Meta, Amazon, ByteDance, Oracle Cloud, AT&T, Verizon, Deutsche Telekom, NTT.

EntityPositionLWLG linkNotes
★ NVIDIA (NASDAQ: NVDA)$7.7B optical-supply-chain investments 2025-2026 (Nokia CPO, Lumentum $2B, Coherent, Marvell); Spectrum-X today → Feynman NVL1152 all-CPO 2028◐ Stage-3 candidate (medium); SVP AI Infra Bjorlin OFC 2026 plenary explicitly framed CPO + low-power optical interconnects as required$1T orders for Blackwell + Vera Rubin through 2027 (per GTC 2026)
GoogleTPU-driven custom-silicon + custom optics; OCS at scale; long-time Coherent + Lumentum customer◐ Indirect — purchases modules from L5Hyperscaler with deepest in-house optics R&D
MicrosoftAzure AI infra; Inphi/Marvell DSP customer; Sirius optical-network research◐ IndirectHeavy optical-supply diversification
MetaOCP-driven 800G/1.6T deployments◐ IndirectOpen Compute Project standards-setter
Amazon (AWS)EC2/Trainium custom-silicon + SiPh in-house◐ IndirectDevelops in-house optics (Annapurna/Trainium2 + SiPh)
ByteDanceMajor Chinese hyperscaler; H100/H200 cluster builder◐ IndirectEoptolink/Innolight customer
Oracle CloudOCI scaling; Coherent + Marvell customer◐ IndirectSmaller hyperscaler, growing fast
AT&T, Verizon, Deutsche Telekom, NTTTelco backhaul + metro coherent◐ IndirectInP + TFLN-led; LWLG less direct here

Choke point at L6:

  • Modulator-material decision authority. Hyperscalers typically buy modules, not modulators. They influence by writing OCP-style specs that filter back through L5 → L4 → L3 → L2 → L1.
  • NVIDIA’s Spectrum-X / Feynman roadmap. This is the single-most-watched L6 spec; whatever modulator material NVIDIA’s optical-engine partners specify for 2027-2028 wins.

2026-2028 outlook for L6:

  • 2026: 1.6T pluggable deployment continues; CPO engineering trials.
  • 2027: 3.2T pluggable + first hyperscaler CPO clusters in production.
  • 2028: NVIDIA Feynman NVL1152 all-CPO; full hyperscaler shift accelerates.

Stack-traversal: where LWLG revenue flows from a single Stage-3 customer

Hypothetical: Coherent Corp Stage-3 design-in for 3.2T transceiver (2027 production target)

Example revenue path — one hypothetical Stage-4 deal

single trace from polymer → AI rack · concrete vendor choices
L1
Lightwave Logic
★ source of value
Perkinamine polymer · priced per-gram royalty + per-wafer license
↓ qualifies the chemistry through K-POD process flow
L2
GlobalFoundries Fotonix
polymer-modulator wafer with Coherent-specific layout
↓ singulation + die-out
L3
Marvell DSP (or Broadcom Taurus)
drives the modulator at <1V swing
↓ pairs in module assembly
L4
Foxconn / Jabil
singulation, fiber-attach, hermetic cap (4th-gen ALD reduces hermeticity cost)
↓ shipped as a packaged optical engine
L5
Coherent Corp
packages as 3.2T XPO transceiver under Coherent's brand
↓ delivers via VAR / direct
L6
NVIDIA / hyperscaler
installs in AI factory rack
Where LWLG earns
Two recurring revenue events per wafer: (a) per-gram polymer sale at L1→L2 transition, plus (b) per-wafer / per-modulator royalty tied to the Coherent-specific PDK at L2. Once the L1↔L2 PDK lock-in is established, every downstream module shipped through L3→L6 reinforces LWLG's recurring stream — the customer cannot re-tape-out at a different foundry without $1-3M / 6-12 months of re-qualification.

LWLG captures L1 economics. The strategic moat is the L1↔L2 PDK lock: once Coherent designs at GF Fotonix using LWLG’s PDK, switching to NLM Selerion or TFLN requires re-spinning the modulator section, which is multi-year work. This is the structural reason Stage-3 is a 12-18 month investment by the customer — and once it converts, switching cost approaches infinite.

Cross-stack choke points (priority watch list)

Choke pointStack layerWhy it matters2026-2028 implication
TSMC COUPE modulator-material decisionL2TSMC partners with NVIDIA on CPO; if TSMC adopts polymer (vs TFLN/BTO), supply curve and LWLG addressable market expand 5×+Most-binary thesis-changing event
NVIDIA Feynman 2028 modulator specL6 → cascades to L5/L4/L3/L2/L1Directly forces the full stack to a single material choice for NVDA’s ~50% of optics demand2027 disclosure window
Marvell direct LWLG MSA (vs Polariton-only)L1 → L3/L5Would convert LWLG from “indirect-via-Polariton” to direct counterparty for the entire Marvell DSP customer baseOpen question post-Apr 22 acquisition
Foundry capacity expansion (GF Fotonix; Tower PH18)L2Wafer-volume bottleneck; 2027-2028 demand may exceed 2026-deployable fab capacityCapex announcements
Eoptolink / Innolight TFLN→polymer pivotL5If the Tier-1 Chinese transceiver duopoly adopts polymer for 3.2T, LWLG volume curve compounds with NVDA ordersWatch OFC 2027 + ECOC 2026/27 disclosures
NLM patent-overlap dispute (latent)L1If LWLG-NLM patent-stack overlap materializes into litigation, slows down bothNo public litigation as of Apr 2026
Engineering-staffing bottleneck at LWLGL1 → L2 PDK onboardingLeMaitre at ECOC 2025: “Several Foundries waiting to be onboarded but LWLG doesn’t have enough engineering resources”Hiring patterns at LWLG visible via LinkedIn

Summary — LWLG’s stack position in one sentence

LWLG is an L1 specialty-chemistry IP holder with PDK-anchored L2 distribution, indirect L3-L5 reach via foundry partners (GF/Tower/SilTerra) and module/DSP customers (Marvell-via-Polariton + 3 unnamed Stage-3 F500), and L6 demand pull driven by NVIDIA’s $7.7B optical-supply-chain commitments — with the Stage-3-to-Stage-4 conversion being the single binary event that converts the IP into recurring royalty revenue.

Sources