Terence S.-Y. Chen
Latitude Design Systems
Abstract
Silicon photonic integrated circuits utilize CMOS processes to achieve high-density photonic integration, an important emerging technology. As chip complexity increases, computer-aided design (CAD) tools are critical to unleash the potential of this technology. This white paper outlines the progress, state-of-the-art, challenges, and opportunities in design automation for silicon photonics ICs. Firstly, we introduce the silicon photonic design flow, distinguishing between device and system levels. We then discuss the adoption of electronic design automation (EDA) techniques like schematic-driven design and hierarchical reuse. This is followed by examining compact modeling and simulation technologies, including variability analysis and photonic-electronic co-simulation. We also detail physical implementation considerations such as manufacturability, optical routing constraints, etc. Finally, we identify gaps in areas like standardization, multifunctional tools, complex effects modeling etc. that require collaborative resolution to achieve scalable and reliable silicon photonic IC CAD capabilities. We also introduce PIC Studio, an integrated silicon photonic IC design platform developed by Latitude Design Automation, comprising modules like PhotoCAD, pSim, Advanced SDL, and pLogic to enable full flow photonic IC design from theory to practice. This widely adopted platform facilitates maturity and prosperity of the silicon photonics industry.
Keywords: Silicon photonics, integrated circuits, design automation, electronic design automation, computer-aided design
1. Introduction
Photonics technology has been around for decades, but in recent years the incorporation of silicon photonics into integrated circuits has gained traction in the semiconductor industry. Why? The historic electronic IC design paradigm faces increasing challenges in meeting goals of faster signal processing, lower power, and smaller footprint. Progressing to smaller process nodes is becoming more difficult, taking longer and costing more. Photonics in general and silicon photonics in particular, represents technologies that can help overcome these challenges.
In the early 1980s, ICs were designed primarily by PhDs with advanced knowledge of device physics, and manufactured in corporate fabs by engineers skilled in materials processing. The focus was on creating the best (fastest, lowest power, smallest) transistors through layout by hand and extensive device simulation. This design style was laborious, time-consuming and prone to errors requiring expensive respins.
In contrast, the electronic IC industry today designs and manufactures chips with billions of transistors and interconnects, keeping design-to-manufacturing time and costs under control. This transformation is enabled by innovations in organizational structure and automation.
The field of electronic design automation (EDA) has enabled remarkable advances in integrated circuit design over the past five decades. EDA refers to the use of computer-aided design tools to design, simulate, and verify the functionality of electronic circuits before fabrication. As integrated circuits have increased in complexity, from the first microprocessors in the 1970s to modern multicore CPUs and GPUs, EDA tools have been critical in managing this design complexity and enabling more efficient workflows.
Several key factors have driven the advancement of EDA: manufacturing technology, computing power, design standards, design methodology, and the EDA tools themselves. Smaller transistor geometries, made possible by improvements in semiconductor fabrication plants, allow for greater numbers of transistors per chip. This has enabled integration of full systems onto single chips. More powerful computers have allowed EDA software to handle the complexity of simulating and verifying billion-transistor designs quickly and accurately.
Standard design abstractions and methodologies like register-transfer level (RTL) design have enabled reuse and productivity gains. And continuous advancement of EDA tools for tasks like logic synthesis, physical design, and formal verification have increased automation and eased the designer's burden. The complexity growth in integrated circuits over the past five decades has been astonishing. The first microprocessor, Intel's 4004 released in 1971, contained 2,300 transistors and was manufactured using a 10 micron process. Today's advanced CPUs like the Apple M2, released in 2022, use a 5nm manufacturing process to integrate over 20 billion transistors onto a single chip. The M2 also integrates eight high-performance CPU cores along with a 16-core neural engine and 10 graphics cores, representing a massive integration of heterogeneous computing resources.
Such integration would not be possible without the productivity and complexity-management capabilities provided by electronic design automation. EDA has been pivotal in enabling the integration of entire systems onto tiny slivers of silicon, fulfilling the promise of Moore's Law and fueling the computing revolution. The rapid advancement of EDA and manufacturing technologies will continue to accelerate innovation in the semiconductor and computer industries.
It began with the separation of IC design and manufacturing, leading to dedicated foundries and fabless design companies. By the 1980s, the foundry-fabless ecosystem was firmly established. In 1987, TSMC pioneered the pure-play foundry model, enabling easy market access for fabless companies.
Silicon photonic IC design automation is at a similar nascent stage today, with some gaps compared to mature EDA. But with collaborative resolution of these challenges, comparable design efficiency and reliability can potentially be achieved.
2. Silicon Photonic Design Flow
The silicon photonic IC design flow can be broadly divided into device/component design and system/circuit design.
2.1 Device and Component Design
This involves designing and optimizing layouts and physical parameters of photonic building blocks like waveguides, resonators, modulators etc. The focus is tailoring optical characteristics by engineering component geometry and layout. Simulation-driven optimization is performed using EM field solvers like FDTD, FEM, EME etc. Objective functions can include transmission, extinction ratio, Q-factor, sensitivity etc. Manufacturing variability can be modeled through 'virtual fabrication'. Robust optimization can help create fabrication-tolerant components. Though computationally intensive, such EM-level design is critical for high-performance photonics.
2.2 System and Circuit Design
At this higher level of abstraction, the aim is to logically connect components to implement system functionality. A key consideration is the design approach:
Layout-driven: The physical integrated circuit layout is directly drawn by the designer, incorporating devices to meet specifications. This enables early control over device physics.
Schematic-driven: The system is captured hierarchically using an abstract description, from which layout is automatically synthesized. This resembles EDA flows.
In layout-driven design, also termed 'physical synthesis', designers directly draw the photonic circuit layout, deciding component placement and optical connections to meet design goals. For smaller circuits, this can provide efficient design by precise control over actual device geometry. However, the lack of hierarchy, abstraction and automation makes such approaches harder to scale.
Therefore, designers are increasingly adopting schematic-driven design conceptually similar to electronic EDA. Here the required photonic circuit architecture is captured using a schematic editor, with abstract symbols for components from a library. Connectivity is indicated in this schematic, which serves as the architectural specification. Physical layout is automatically synthesized from this hierarchical schematic while maintaining connectivity. Schematic-driven design provides major advantages in managing complexity through hierarchy, reuse and layout automation. High-level schematics can also be simulated using compact device models prior to physical implementation. Integration with electronics at the schematic level also becomes straightforward. While layout-driven techniques continue finding prominence for smaller custom circuits, schematic-driven design has emerged as the leading methodology for complex photonic ICs, similar to electronic design automation (EDA).
In summary, a combination of physics-driven and system-driven approaches spanning nano to microscale is indispensable to implement integrated electro-optic systems. As complexity scales, automated schematic-driven system design tightly integrated with lower-level EM layout and simulation is becoming critical to leverage the potential of silicon photonics.
3. Adopting EDA Techniques
Electronic design automation refers to software tools and methodologies for electronic IC design, which has been crucial for the tremendous complexity scaling over the past decades, enabling reliable automation for billion-transistor chips. As complexity rises, adopting EDA techniques becomes highly imperative for silicon photonics design. Some key EDA paradigms embraced are:
3.1 Schematic Capture
The photonic circuit architecture is captured using a schematic editor, with abstract symbols representing components like waveguides, couplers, modulators etc. Connectivity between components is indicated in the schematic. This conceptual schematic or 'netlist' enables functional simulation and drives physical layout synthesis. It captures the system specifications in a technology-independent manner.
3.2 Hierarchical Reuse
The circuit schematic is constructed hierarchically from lower-level blocks and sub-circuits. Common subsystem functions are modularized and designed once for reuse. This avoids redundant redesign while enabling complexity scaling. The blocks can be represented as 'black-boxes' allowing abstraction.
3.3 Interfacing with Electronics
Most applications require integrating photonics with electronics. The same schematic editor is used to capture photonic and electronic sub-systems, enabling co-simulation and a unified flow.
3.4 Layout Synthesis
Parameterized layout generator cells called PCells automate rendering the circuit layout. Connectivity and component choices from the high-level schematic drive physical layout assembly. Hierarchical layout realization enhances reuse.
3.5 Design Verification
Steps like design rule checking (DRC) for manufacturability validation and layout-versus-schematic (LVS) comparison for functional correctness provide critical signoff before fabrication to reduce costly respins.
3.6 Process Design Kit
Foundry-provided libraries with verified parametric cells, abstract symbols, compact models, layout rules, layers, resolutions etc. enable process-specific realization and reuse.
These EDA techniques allow managing complexity through hierarchy, abstraction and reuse while enhancing productivity and reducing errors. However, directly applying electronic design techniques to photonics without customization is also insufficient. Photonics poses unfamiliar constraints like optical routing restrictions, layout sensitivity, complex curvilinear geometries, compact modeling challenges etc. Therefore, while adapting EDA paradigms is a key driving force as complexity rises, specialized solutions augmenting conventional EDA capabilities are indispensable for silicon photonics.
4. Compact Modeling and Circuit Simulation
4.1 Need for Compact Models
Circuit simulation plays a vital role in the schematic-driven design flow by predicting integrated system behavior without fabrication. However, optical simulation of large photonic circuits requires compact parameterized models of individual components that can accurately capture device physics while being computationally efficient compared to full-wave electromagnetic simulators. Developing such robust, validated silicon photonic compact models remains an active research area.
4.2 Compact Model Development
Compact models approximate complex optical phenomena like dispersion, propagation loss, nonlinear effects using analytical equations and circuit equivalents with extracted device parameters. The detailed physical layout is abstracted into a set of lumped parameters encapsulating performance. Generating accurate, scalable compact models requires extensive device simulation and experimental characterization for parameter extraction. Lack of standardization poses interoperability issues in model interchange between different simulation tools.
4.3 Co-Simulation of Electronics and Photonics
Many applications like microwave photonics, sensors, optical interconnects require concurrent simulation of electronics and photonics. This necessitates integrating compatible parameterized behavioral models of electro-optic components, and simulator engines capable of handling both domains. Joint simulation reveals interactions between electronics and photonics while guiding co-design.
4.4 Variability Analysis
Photonic performance varies substantially with nanometer-scale statistical manufacturing variations. Compact model libraries must encapsulate these effects to enable yield analysis through Monte Carlo simulations over a large ensemble of parameter variations. Spatial correlations between adjacent components also require modeling.
The complexity of optical phenomena coupled with multiphysics interactions makes compact modeling far more challenging compared to electronics. Implementing optical effects like nonlinearity and noise on electronic simulators also remains difficult. Foundry involvement in providing standardized verified PDKs with comprehensive model libraries would be transformative. Mature techniques for statistical analysis, design centering and optimization considering manufacturing variability in electronics and photonics also need development to enable predictive yield-driven design. Addressing compact modeling and multiphysics co-simulation challenges will likely be the most critical enabler for predictive and reliable photonic IC CAD.
5. Physical Implementation and Manufacturing
While the schematic-driven design flow provides benefits of hierarchy and abstraction, the photonic circuit must ultimately translate to a tangible manufactured layout. This physical implementation involves certain unique considerations:
5.1 Curvilinear Layout
Photonic waveguides involve smooth curves to minimize optical loss, necessitating true curvilinear geometries, unlike rectilinear electronics. This poses layout generation and verification challenges.
5.2 Routing Constraints
Photonic connections are restricted to a few routing layers unlike multilayer electronics. Waveguide crossings induce optical loss and reflections. Optical routing congestion typically requires human intervention. These factors significantly complicate scalable automated routing of large photonic ICs.
5.3 Layout Density Constraints
A uniformly high pattern density is required across the layout to avoid lithography artifacts. Empty areas must be filled with non-functional ‘dummy’ geometries to meet foundry requirements. This consumes design space while complicating routing.
5.4 Technology Files and PDKs
Foundries provide technology files with critical information like waveguide widths, resolution, bend radii, layer mappings etc. for PCells. A qualified PDK contains verified parametric cells, layout rules, and models. Immature PDKs pose adoption barriers.
5.5 Design Rule Checking
Foundry design rules for minimum features sizes, spacings, angles etc. must be obeyed for manufacturability. DRC verifies layout compliance to enable fabrication. Custom DRC rules suited for photonic geometries are essential.
5.6 Test Structures
Specialized structures to characterize components and extract simulation models must be designed alongside the main circuit. PDKs incorporate standardized test cells.
These physical implementation considerations make layout design and verification extremely important for photonic ICs compared to digital electronics flows. Designers must navigate layout constraints arising from optical physics while meeting stringent fabrication requirements. Tools must become lithography and variation-aware. Addressing constraints emerging from physical implementation and manufacturing compatibility is vital to harness the fabrication capabilities offered by silicon photonics.
6. Remaining Gaps and Opportunities
While silicon photonic design automation has adopted electronic design concepts, some distinctions still remain. Resolving these gaps can help make photonic IC development truly scalable, predictable and amenable to automated synthesis:
6.1 Lack of Standards
The lack of common standards for schematic symbol representations, compact model interfaces, and design interchange formats hampers interoperability, IP reuse and tool integration. Defining photonic equivalents of established electronics standards like Verilog-A or SPICE could accelerate progress.
6.2 Need for Multifunctional Tools
Dedicated tools tailored for photonics design automation that holistically incorporate layout, variability-aware simulation, verification, manufacturing-compliance, electronics co-design etc. with high productivity are still absent but urgently needed.
6.3 Complex Optical Effects Modeling
Modeling complex phenomena like nonlinearity, dispersion, noise, and frequency-dependence for multi-carrier optical signals remains quite difficult on traditional EDA tools optimized for electronics. Specialized techniques are required.
6.4 Statistical Analysis for Manufacturing Yield
Mature techniques for automated statistical analysis, design centering, and optimization considering correlated manufacturing variability in both electronics and photonics need development to enable predictive yield-driven design.
6.5 Lack of Qualified PDKs
Foundry-validated PDKs with complete component libraries containing verified parametric layouts, robust models across domains, and comprehensive rule decks are still rare. Close foundry involvement can greatly facilitate component reuse.
6.6 Packaging and Fiber Interface Design
Tool support for system-level considerations like designing fiber coupling structures, electrical contacts, and links to optical/electronic packages is needed. Automated techniques can alleviate packaging constraints.
Addressing these gaps presents research opportunities and underscores the need for collaboration between academia, EDA vendors, process developers, system integrators and end-users to advance development. Creating a comprehensive silicon photonics design ecosystem encompassing standards, libraries, interoperable tools, and mature PDKs promises to significantly enhance design productivity, performance and reliability for this emerging technology.
7. PIC Studio Design Platform
Latitude Design Automation has independently developed the PIC Studio silicon photonic IC design platform, comprising modules like PhotoCAD, pSim, Advanced SDL, and pLogic to enable full-flow photonic IC design from theory to practice.
7.1 PhotoCAD Layout Design
PhotoCAD is a dedicated tool for photonic IC layout design. It supports true curvilinear waveguide layouts and provides intelligent optical routing and component placement. Designers can quickly draw layouts and iterate with post-layout simulation for optimization.
7.2 pSim Simulation Engine
pSim is a photonic circuit simulator for time-domain and frequency-domain simulations. It supports compact models, enabling functional verification, parameter analyses of photonic circuits. Users can conveniently import and simulate various devices and subsystems.
7.3 Advanced SDL
Advanced SDL enables automatic conversion from schematic to layout. It allows designers to operate at an abstract level and automatically generates physical layout from schematic connectivity without manual drawing.
7.4 pLogic Photonic Integration
pLogic provides a library of photonic logic gates to construct digital photonic circuits. Designers can leverage photonic logic gates to build various digital functionalities, similar to using electronic gates.
The PIC Studio platform is widely adopted to complete design from device to system levels. It enhances design efficiency, reduces time-to-market, and facilitates the rapid growth of silicon photonic ICs. Latitude Design Automation will continue innovating to further augment the capabilities of this platform.
8. Summary and Outlook
This white paper has outlined the evolution, current state, challenges, and future directions for design automation of silicon photonic integrated circuits. Advancing design platforms is imperative to support integration scale-up of this technology. Adopting EDA flows and tools is the key enabler towards this goal. Compared to mature electronic IC design automation, some gaps persist in silicon photonics automation like lack of standards, modeling complexity etc. which require collaborative resolution across industry, academia and research. We have also introduced Latitude Design Automation's PIC Studio design platform, whose comprehensive capabilities support the entire silicon photonic IC design cycle. Adoption of this platform strongly facilitates the maturation and prosperity of silicon photonics technology. We believe that with persistent innovation, silicon photonic IC design automation will become a vital infrastructure driving photonics-electronics convergence and enabling advanced applications.
Kommentare