返回搜尋
穎崴 2026Q2 法人說明會
6515上市
法人說明會
穎崴 2026Q2 法說會簡報重點與營運摘要

WinWay Technology 法說會簡報

Company Overview

WinWay Technology (穎崴科技) is a trusted partner in semiconductor testing.

Products & Technologies

The Evolution of Co-Packaged Optics (CPO) Advanced Testing Methodologies for Silicon Photonics

  • CPO (Co-Packaged Optics): An advanced packaging technology integrating optical engines directly with ASICs, reducing distance and improving performance.
    • Benefits:
      • Speed Evolution: From Gbps to Tbps (800G to 1.6T ~ 6.4T).
      • Energy Efficiency Leap: Saving Over 50% Power Consumption (20 pJ/bit to below 5 pJ/bit).
      • Spatial Optimization: Chip-level Packaging Revolution (reducing latency by over 15%).
    • Architecture: Consists of Optical Engine (PIC, EIC), HBM, ASIC, Interposer, Substrate, PCB, Fiber Array Unit (FAU), Fiber, and External Laser Source.
  • CPC (Co-Packaged Copper): An alternative co-packaging approach using copper interconnects.
    • Characteristics: Longer electrical Path (cm level), Hot-swappable & field replaceable, Higher power consumption, Mature ecosystem & cost-effective solution.
  • Photonic Integrated Circuit (PIC) Blueprint:
    • Tx Path: Source (Light Generation) → Modulator (Data Writing) → Routing (Waveguides) → WDM (Multiplexing) → Coupler (Off-Chip) → Fiber.
    • Rx Path: Fiber → Coupler (On-Chip) → Routing → WDM (Demultiplexing) → Photodetector (Conversion) → Electrical Output.
    • ASIC Data drives the Modulator Driver.
  • Light Source Matrix (Step 1):
    • DFB Laser (Distributed Feedback): Highly Stable wavelength, narrow linewidth. Ideal for standard 1310nm/1550nm telecom.
    • VCSEL (Vertical-Cavity Surface-Emitting Laser): Low cost, easy to form arrays. Ideal for short-reach data centers. Limitation: Less suited for long-reach / high-power.
    • Comb Laser: Crucial for 102.4T CPO. Generates multiple precise wavelengths from one device, enabling massive WDM density.
  • Modulator Showdown (Step 2):
    • MZM (Mach-Zehnder):
      • Footprint: Large (1-3 mm)
      • Bandwidth: High
      • Thermal Stability: High
      • PAM4 Linearity: Excellent
      • Yield / Maturity: Highest (10+ Yrs)
      • Customers: Broadcom / Intel / Marvell / Lightmatter / Cisco
    • MRM (Micro-Ring):
      • Footprint: Micro (10-20 um)
      • Bandwidth: Medium
      • Thermal Stability: Critical (Requires TEC)
      • PAM4 Linearity: Moderate
      • Yield / Maturity: Medium
      • Customers: nVIDIA / AyarLabs
    • EAM (Electro-Absorption):
      • Footprint: Small (50-100 um)
      • Bandwidth: Medium
      • Thermal Stability: Medium
      • PAM4 Linearity: Poor
      • Yield / Maturity: Medium
      • Customers: Coherent / Lumentum / Intel
  • Routing The Light (Step 3): Utilizes waveguides with Total Internal Reflection (TIR).
    • Power Splitters (MMI: Multimode Interference): Input Light (Cyan) → MMI Mixing Region → 4x Equal Output Light (Cyan).
  • Wavelength Division Multiplexing (WDM) (Step 4): Uses a prism to multiplex different wavelengths (e.g., 1270 nm to 1410 nm).
    • Target KPI: Crosstalk must be maintained at -20 dB.
  • I/O Coupling Strategy (Step 5):
    • Grating Couplers: Light enters/exits vertically. High insertion loss, wavelength sensitive. Advantage: Enables wafer-level testing (testing from the top before dicing).
    • Edge Couplers: High bandwidth, ultra-low insertion loss (<1.0 dB). Polarization insensitive. Ideal for 128-core FAU. Challenge: Demands extreme sub-micron alignment precision.
  • Optical Engine Design:
    • 2D: Wirebond (EIC, PIC, Substrate)
    • 2.5D: EIC, PIC, Interposer, Substrate
    • 3D-WB/FC: Wirebond (EIC, PIC)
    • 3D Hybrid integration -TSV or Fan-Out: Hybrid Bonding (EIC, PIC, Interposer, Substrate)
    • 3D Hybrid integration -Hybrid bonding: Hybrid Bonding, TSV (EIC, PIC, Interposer, Substrate)
    • 3D-Monolithic: EPIC, Substrate
    • Optical Engine Connector examples: CORNING, Teramount, SENKO.
  • Single Mode Fiber (SMF): The only choice for CPO due to silicon photonics waveguide being single-mode.
    • Benefits: Supports only 1 fundamental mode, Maintains signal over 2 km+, Native match with SiPh waveguide.
    • Result: Enables 200 Gbps+ over 2 km+, ideal for CPO deployment.
  • Active Alignment: Critical for FAU.
    • Challenges: Fiber Roundness (≤0.8μm), Core Concentricity (≤0.5µm), Fiber Size (±0.71.0μm), V-groove Angle (±0.3°), V-groove Depth (±0.51.0μm), V-groove Pitch (±0.3~0.5µm).
    • Tolerance Stack-up: 3.8µm | 42% of core diameter.
    • Tools: Multi-Axis Photonics Alignment System, Hexapod, Piezo System for Nano positioning and Fiber Alignment.
  • CPO Production Assessment:
    • Solution 1 Golden FAU (Fixed Reference Focusing): Establish Golden Device → Establish Golden FAU → Move All FAUs to Reference Position → Verify Focusing Efficiency.
    • Solution 2 Self-align (Auto Focus Calibration): Measure Intrinsic Component Variation → Build Auto Calibration Model → Auto Focus Calibration → Verify Focusing Efficiency.
  • CPO Production Bottleneck:
    • Precision Packaging & Lasers: Sub-micron alignment.
    • Lack of Standardization: Different form factors (NVIDIA, QSFP, Intel).
    • Maintenance & Serviceability: External Laser Module, Downtime ↑ Cost.
    • Yield & Cost Risk: High costs, potential low yield.
  • Co-Packaged Optics Test Flow:
    • Wafer Level → Die Level:
      • Components: EIC/PIC Wafer, EIC/PIC Die.
      • Testing: WLCSP Fine Pitch Probe Head (EIC/PIC Wafer test).
      • Methodology: Electrical at TOP / Optical at BTM. Electrical Test Instrument (DC / RF / BERT), Optical Test Instrument (Laser / OSA / Power Meter).
    • Package Level:
      • Components: Optical Engine, CPO Substrate.
      • Testing: Optical and Electrical Test Socket (Optical Engine), Double Sided Probing System.
      • Methodology: Active alignment, Passive alignment, Direct with FAU.
    • Module Level:
      • Components: ASIC, PIC, EIC, OE, CPO Module.
      • Testing: HyperSocket™.
      • Methodology: Pick & Place, Place FAUs, Plug All FAUs, Plug & Play.
  • WinWay's Innovative Test Solution: HyperSocket™:
    • Requirements: Package Size > 100 X 100 mm², Pin count > 10,000 pin, Device Warpage > 0.4 mm, High Current Density > 6 A, SACQ (Frequent Clean) ~10 TD, Solder Ball Melting, Probe bent and worn out issue.
    • Advantages over Conventional Test Interface (Elastomer/Spring probe):
      • Significantly increased contact surface.
      • Less and stable contact resistance (↓30%).
      • Higher current carrying capacity (↑30%).
      • Less joule heat generation (↓30%).
      • No housing warpage of spring probe socket.
      • Lower total test cost of ownership (Factory operation, Hardware, Maintenance, Test interface including Change Over Kit, Socket, PCB).
    • HyperSocket™ Models:
      • Hyper-UF: For Frequent Clean, Solder Ball Melting.
      • Hyper-LF: For Ultra Large Package, High Pin Count.
      • Hyper-DH: For High Current Density, Probe Bent.
      • Hyper-Liquid: Under Validation, for Extremely High Power (> 2500W).
    • Patent NO.: TWI862047, TWI922268, TWI901181, TWI862191, TWI884802, CN220584352, TWI901161, TWI923382, US(Granted).

Clients & Markets

Optical Industry M&A Wave in the AI Era

  • Nov 2024: Celestial AI acquires Rockley Photonics IP ($20M, SiPho IP consolidation).
  • Dec 2025: Marvell to acquire Celestial AI ($5.5M, Optical scale-up for AI).
  • Feb 2025: Nokia acquires Infinera ($2.3B, Optical networking & DCI).
  • Apr 2026: Credo to acquire DustPhotonics ($1.3B, SiPho PIC vertical integration).
  • May 2025: AMD acquires Enosemi (Undisclosed, SiPho PIC for CPO).
  • Apr 2026: Marvell acquires Polariton Technologies (Undisclosed, Plasmonics + SiPho modulation).
  • Total M&A value: ~$10B+.

Chipmakers & Hyperscalers Back Photonics

  • Oct 2024: Google Ventures et al. → Lightmatter ($400M, CSP → Photonic Computing).
  • Feb 2026: MediaTek → AyarLabs ($90M, IC Design → Optical I/O).
  • Oct 2024: Cisco · NVIDIA → Xscape Photonics ($44M, CSP/Networking + IC → DWDM).
  • Mar 2026: NVIDIA → Lumentum ($2.0B, IC Design → Laser / SiPho).
  • Dec 2024: AMD · Intel · NVIDIA → AyarLabs ($155M, IC Design (joint) → Optical I/O).
  • Mar 2026: NVIDIA → Coherent ($2.0B, IC Design → Laser / Networking).
  • Total investment value: ~$4.7B+.

Silicon Photonics Supply Chain

  • CSP (End Customer): NVIDIA, AWS, Microsoft, Meta, ByteDance.
  • SOI Wafer (Photonic Wafer): soitec, ShinEtsu, SIMGUI, Intelli EPI.
  • Connector: TAN, SCONEC, SENKO, samtec, SIEMON, Amphenol, 睐芯科技, LSI LightSense.
  • Testing: WinWay, MPI Corporation, HON PREC, Chroma.
  • PIC Design & EDA: IBM, intel, sicoya, MACOM, Juniper, CISCO, SiFotonics, POET, Ansys, AyarLabs, FUJITSU, MARVELL, LUMENTUM, Infinera, HUAWEI, RANOVUS, cadence, SKORPIOS, OpenLight, SYNOPSYS.
  • Foundry: intel, Tower, A silex, GlobalFoundries, SAMSUNG UMC, ST, Acacia.
  • EIC & DSP Design: BROADCOM, COHERENT, LUMENTUM, MACOM, ciena, Infinera.
  • Light Source (from EPI wafer): LandMark, COHERENT, SUMITOMO ELECTRIC, Intelli EPI, IQE, NICHIA.
  • Fiber: CORNING, YOFC, SUMITOMO ELECTRIC, FURUKAWA ELECTRIC, 富通集团, FUTONG GROUP, FiberHome, prysmian, ZTT, Fujikura.
  • ELS: FURUKAWA ELECTRIC, BROADCOM, LUMENTUM, COHERENT, SUMITOMO ELECTRIC, TE, O-Net Technologies.

Outlook & Strategy

Co-Packaged Optics Trend

  • ~2016: Pluggable Optics (100% Cu)
  • 2023: OBO (On-Board Optics) (80% Cu + 20% Optics)
  • 2025: NPO (Near-Packaged Optics) (50% Cu + 50% Optics)
  • 2027: 2.5D CPO (COUPE Switch) (20% Cu + 80% Optics)
  • 2030~: 3D CPO (COUPE XPU) (100% Optics)
  • Key Metrics Evolution:
    • Bandwidth: 800G → <1.6T → <3.2T → <6.4T → 12.8T
    • Power Use: 1x → ~0.8x → ~0.6x → <0.5x → <0.1x
    • Latency: 1x → <0.75x → <0.5x → <0.1x → <0.05x
  • Source: Counterpoint Research《矽光子與共同封裝光學(CPO)報告》

Key Foundries of Silicon Photonics and CPO (2025-2029 Roadmap)

  • TSMC:
    • 2025: Mass Production (Platform: COUPE, PIC Node: 65nm, EIC Node: 7nm or below, Package: SoIC-X).
  • SAMSUNG:
    • 2026: Platform: Samsung Electronics SiPH, PIC Node: 300mm SOI, EIC Node: SF4U/SF2Z, Package: Vertical Integration.
  • Intel:
    • 2026: CPO (Optical I/O), Platform: OCI, PIC Node: Dedicated SiPH Node, EIC Node: Advanced CMOS, Package: 2.5D/3D Packaging.
    • 2029: TBD.
  • GLOBAL FOUNDRIES:
    • 2025: Mass Production (Silicon Photonics Process, Platform: Fotonix, PIC Node: 45nm RF-SOI, EIC Node: EIC/M RF-SOI (EIC/PIC Monolithic), Package: Monolithic).
  • UMC:
    • 2026: Dev./Validation.
    • 2027: Risk Pilot (Silicon Photonics Process, Platform: iSiPP300, PIC Node: 300mm SOI (IMEC Licensed), EIC Node: Paired with Logic Foundry, Package: Heterogeneous Integration).
  • Source: DIGITIMES, 2026/04

CPO Roadmap

  • Bandwidth Progression: 800G → 1.6T/3.2T → 6.4T/12.8T.
  • Packaging Evolution: Packaging → Pluggable → CPO Switch → XPU Optical IO.
  • Optical Engine Evolution: OE (Optical Engine) → 100G MZM → 200G MZM/MRM → 400G CWDM/DWDM.
  • Fiber/FAU Evolution: 8X → 16X → 32X → 64X → 160X.
  • XPU Optical IO Components: Optical Engine (OE), Switch, Substrate, XPU HBM, Interposer, 160X fiber links.
  • Source: TSMC

The I/O Interface Evolution

  • The Past: Electrical Interface.
  • The Present: Electro-Optical Interface.
  • The Future (102.4T+): Co-Design Architecture (Optical Alignment + Thermal Isolation + Signal Integrity).

Future Technical Challenge

  • A. Mechanical Extremes:
    • Package Size > 100 mm up to 200mm.
    • Pin Count > 10,000 pins up to 50,000 pins.
    • Key Challenge: Warpage up to 0.6mm.
  • B. Electrical Performance Barriers:
    • Signal Speed 224Gbps PAM4 and beyond.
    • Key Challenge: Signal Integrity, Crosstalk.
  • C. Thermal Density Crisis:
    • Power Consumption > 4000W per device.
    • Key Challenge: Thermal Runaway / Solder Melt.
  • Technical Development Direction to “Optics": Evolution from mechanical machining accuracy to multi-physics coupled design, encompassing mechanical integrity, electrical performance, thermal management.

Heterogeneous Integration: CPO is the path to success in near future

  • Problem with Organic Substrates: Thermal / Mechanical Instability, Electrical Loss, High CTE Mismatch Silicon (Causes Warping & Misalignment), High Dielectric Loss.
  • Glass as Breakthrough Solution: Superior Stability, Pristine Electrical Performance, Advanced Integration, Near-Perfect CTE Match to Silicon (Ensures Sub-Micron Alignment).

CPO Market Trends & Challenges

  • Market Trends (2025-2030):
    • 2025: nVIDIA: First 1.6T CPO switch.
    • 2026: Spectrum-X(3.2T) mass production.
    • 2027: Optical - Copper coexist.
    • 2028: Scale-up expansion.
    • 2029: Ecosystem maturity.
    • 2030: CPO mainstream.
  • Golden Window (2026-2028): For ASIC platform flexibility. One ASIC supporting both CPO and CPC —Scale-up via CPO for performance, Scale-out via pluggable for flexibility.
  • NVIDIA SIGNALS:
    • 1.6T CPO saves 180MW at million-GPU scale.
    • Spectrum-X: 5x efficiency, 2Tb/s, 10x reliability.
    • CPO and copper coexist, layered deployment.
  • FUTURE TRENDS:
    • Scale-up is the main battlefield.
    • Adoption: coexist → expand → mainstream.
    • ELS architecture, DSP-free design.
  • CORE CHALLENGES:
    • Thermal & yield: one defect kills package.
    • Serviceability: operations SOP rebuilt.
    • Standards & cost: ecosystem immature.

WinWay is Ready for CPO & CPC Testing

  • Offers advanced package test solutions for both Co-Packaged Copper (CPC) and Co-Packaged Optics (CPO) architectures, covering HBM, ASIC, Interposer, Substrate, and PCB.

Additional Data

GPU Architecture AI Chip Roadmap (2026-2035)

  • GPU Die Size: 728 mm² (2026) → 750 mm² (2029) → 700 mm² (2032) → 600 mm² (2035).
  • GPU Power: 800 W (2026) → 900 W (2029) → 1,000 W (2032) → 1,200 W (2035).
  • # of GPU Dies: X2 (2026) → X4 (2029) → X4 (2032) → X8 (2035).
  • # of HBM Stack: HBM4 X8 (2026) → HBM5 X8 (2029) → HBM6 X16 (2032) → HBM7 X32 (2035).
  • HMB Power per Stack: 75 W (2026) → 100 W (2029) → 120 W (2032) → 180 W (2035).
  • Interposer Die Stack: 2,198 mm² (2026) → 4,788 mm² (2029) → 6,014 mm² (2032) → 9,245 mm² (2035).
  • Total Bandwidth: 16/32 TB/s (2026) → 48 TB/s (2029) → 128/256 TB/s (2032) → 1,024 TB/s (2035).
  • Total HBM Capacity: 288/384 GB (2026) → 400/500 GB (2029) → 1,536/1,920 GB (2032) → 5,120/6,144 GB (2035).
  • Total Power: 2,200 W (2026) → 4,400 W (2029) → 5,920 W (2032) → 15,360 W (2035).
  • Chip heat density design reference: 1.1~2 W/mm².
  • Source: KAIST TERALAB

The 448G Electrical Wall

  • Copper Channel Loss: Insertion loss approaches -10dB on just 1 inch of premium dielectric at 112 GHz Nyquist.
  • The Skin Effect Limit: At > 100GHz, skin depth drops below 0.2 um.
  • Dielectric Loss Multiplier: Loss scales linearly with frequency.
  • DSP Power Explosion: Compensating for physical loss requires ADC sampling > 224 GS/s.

Contact Information

🤖 FinmoAI

BETA
AI 驅動 • 法說會分析
法說會逐字稿
生成重點摘要和投資要點
挖掘潛在投資機會
體驗 AI 智能分析

📱 即時通知服務

加入我們的 Telegram 機器人 @diveinvest_bot,每晚 8 點自動推送最新法說會簡報。

加入 Telegram 通知