Poet Technologies Inc. · NASDAQ:POET
Poet Technologies — icOPEN Conference, Singapore (November 2023)
20 slides from Poet Technologies's conference presentation, Nov 30, 2023
Slide 1: Disruptive Photonics Technologies for a New Generation of AI and Data Center Communications
November 30, 2023 Disruptive Photonics Technologies for a New Generation of Artificial Intelligence and Data Center Communications Dr. Suresh Venkatesan POET Technologies
Slide 2: Shift from Trillion Edge Devices added in the next 15 years
Shift from 1 Trillion Edge General Explosion of Devices Purpose to Data added in the workload generation specific next 15 years 5X Increase in the next 5 years computing (Source: ARM) 90% Generated by Machines High Bandwidth , High Speed Connectivity to move data between the Edge and the Cloud
Slide 3: Industry Mega Trends support Huge Growth of Photonics
Industry Mega Trends support Huge Growth of Photonics Photonics 1.0 Photonics 2.0 Photonics 3.0 1980-2000 : Birth of the Internet 2005-2025 : Web2.0 / Social Media / 3D Sensing 2025+ : Entering the Future !! INTRODUCTION OF PROFILERATION OF GROWTH OF Trans-oceanic Telecommunications Cloud Computing
Slide 4: AI: The killer app for Photonics 3.0
“AI (Artificial Intelligence)” : The killer app for Photonics3.0 The unique advantages of Optical Communications AI Training places significant demands on latency and power • Need to lower Power and reduce latency • Disaggregation of GPU / Memory Lightcounting: AI will add $17B in optical transceive
Slide 5: Conventional Data Center Architectures
Conventional Data Center Architectures Coherent / WDM DCI DAC/ AOC Serial Links Cisco Architectures and Data Workloads increasingly supporting ”short reach” Telecom, commercial from 1990s communications Datacom, Telecom, commercial from 1990s 2000s Ultra Short Reach, commercial from ??
Slide 6: Data Center Disaggregation
Data Center Disaggregation Conventional Architecture : A server blade is a “base” unit Disaggregated Architecture : Composable resources Optical Links between Processor and Memory With the conventional processor-centric computing architecture and copper interconnects, the state-of-the-art chips base
Slide 7: The case for INTEGRATION in Photonics
The case for INTEGRATION in Photonics Incumbent technologies are not scalable for applications needing 100’s of millions and billions of units per year COMPLETE Millions/Year INCUMBENT VECTOR PHOTONICS TECHNOLOGIES* INTEGRATION Unit Volume ✘ ✓ 100s of Millions/Year Size/Channel Count ✘ ✓ Cost ✘ ✓ Bi
Slide 8: Photonics Integration Approaches
Photonics Integration Approaches Si Photonics (external Heterogenous Epitaxial Growth on Production Costs InP Hybrid Integration laser) Integration on SOI Silicon III-V Substrate ✔︎ ✔︎ ✔︎ ✔︎ III-V Growth ✔︎ ✔︎ ✔︎ ✔︎ ✔︎ SOI substrate ✔︎ ✔︎ Silicon substrate ✔︎ ✔︎ Assembly Costs Full Active Alignment
Slide 9: Silicon Photonics OR Silicon for Photonics?
Silicon Photonics OR Silicon for Photonics ? Conventional Silicon Photonics has been disappointingly and agonizingly slow to penetrate DCI (Pluggables) • Only a transmit solution (for most part) • Lossy Modulator – marginal performance at 100G/lane • Presence PRIMARILY in parallel optics – not in WD
Slide 10: "Semiconductorizing" Photonics: The POET way
“Semiconductorizing” Photonics : The POET way POET does for Photonics what Integrated circuits did for electronics Semiconductor Discrete Components Integrated Circuit Moore’s Law Electronics • Smaller, Faster, Cheaper, Volume • Transformed the industry – pervasive presence of semiconductors • Trill
Slide 11: "Semiconductorizing" Photonics
“Semiconductorizing” Photonics Wafer scale assembly. Passive alignments. Eliminate wire bonds. Wafer Level Packaging & Passive Alignments Eliminate Wire bonds Testing • Use automated pick-place • Eliminate active alignments of • Achieve semiconductor type equipment to enable high speed Lasers to len
Slide 12: Silicon Optical Interposer platform
Silicon Optical Interposer platform Two layers of low loss optical interconnects Multiple electrical redistribution layers with low RF insertion loss OPTICALL PLANE ELECTRONICS High throughput visually assisted passive “pick and place” assembly of electronics and photonics ICs and components ELECTRI
Slide 13: POET's Optical Interposer: a flexible Photonics platform
POET’s Optical Interposer™ : a flexible Photonics platform A proprietary technology platform for integrating photonic solutions • Drives down the cost of component POET’s Optical Interposer-based Optical Engines and Light Sources integration and packaging InfinityTM StarlightTM • Improves performanc
Slide 14: Simplified Design, Improved Energy and Lower Cost
Simplified Design, Improved Energy and Lower Cost Data Communications Challenges Conventional Discrete Assembly 22.5 • Serial data communication channels have not been able to keep up with Mux ROSA the pace of bandwidth growth. DSP F A F A Dmux Mux U U • Number of communications lanes F A F A Dmux i
Slide 15: How POET solutions lower the cost of the External Light Source
How POET solutions lower the cost of the External Light Source Interposer approach enables unparalleled performance and cost benefits Incumbent Solutions Interposer Solutions Hybrid Laser Solutions -75% -85% GAIN CHIP Laser Array PM Fiber PLC PM Fiber Array Array DFB Optical Interposer PM Fiber GAIN
Slide 16: 800G Products with Integrated Hybrid Electronics
Interposers enable Flexible Architectures for multiple applications EICs Lasers/PDs EICs PIC Lasers/PDs Optical Interposer with lasers ASIC ASIC Optical Interposer with lasers + SiPh PIC + Detectors PCB PCB Modulated Laser solutions using the Optical Interposer Integrated Laser/PIC solution using th
Slide 17: Extending the path to 3.2T Pluggable Optics
Extending the path to 3.2T Pluggable Optics Solutions scalable to 1.6T and 3.2T architectures with 200G/lane components Highly Scalable Solutions enabling 800G FR4 RECEIVER 3.2Tbps PLUGGABLE OPTICS Support for Externally Modulated Lasers and Drivers Support for Backside Illuminated High Speed Photod
Slide 18: Photonics for Compute
Photonics for Compute Linear Algebra (encoding data as continuous signals) Perfectly suited for Analog compute - Perfectly suited for artificial neural networks, Machine Learning and AI - Manipulates matrices - With appropriate modulating devices, it is possible to encode data using light beams an
Slide 19: In Summary
In Summary • Increasing electrical and optical data rates (electrical – 28Gbps to 224Gbps ; optical – 25G/l Photonics Platform Requirements to 200G/l) • Integration is essential (co packaging of electronic and photonic die on a common • Increasing levels of integration in the module – substrate) wav
Slide 20: Slide 20
POET Technologies — icOPEN Conference, Singapore (November 2023) on 6ix