Co-packaged optics simulation for AI infrastructure

As AI data centres push interconnect bandwidth, power, and integration limits, co-packaged optics (CPO) demands a new generation of multiphysics simulation workflows that unite photonics, electronics, and advanced packaging design.

By Shin-Sung Kim, Sr. Architect, Applications Engineering, Synopsys

Why AI infrastructure is driving the rise of CPO?
The rapid growth of large-scale AI training and inference clusters is transforming the networking requirements placed on data centre infrastructure. As GPUs, accelerators, and switching ASICs continue to increase in performance, the amount of data that must move between devices is accelerating rapidly. Traditional copper interconnects and pluggable optical modules are increasingly challenged by bandwidth density, power consumption, and packaging constraints.

CPO has emerged as one of the most promising solutions to overcome these limits. By integrating photonic and electronic functions within the same package, CPO dramatically shortens electrical interconnect lengths between the switching ASIC and optical engine. This reduces channel loss, lowers power consumption, and enables higher aggregate bandwidth. Studies show that CPO architectures can deliver substantial power savings compared with conventional pluggable optics, while enabling the scaling needed for next-generation AI fabrics.¹

However, achieving these benefits requires far more than simply placing optics closer to electronics. A modern CPO platform combines photonic integrated circuits (PICs), electronic integrated circuits (EICs), optical coupling structures, advanced packaging substrates, thermal management solutions, and high-speed electrical channels. Each influences the others through tightly coupled physical effects. Consequently, multiphysics simulation has become a strategic enabler for AI infrastructure innovation.

Figure 1: Co-packaged optics for AI infrastructure.

Why CPO creates new engineering challenges?
Unlike traditional optical modules, CPO architectures require optimisation across multiple engineering domains simultaneously. Four challenges are particularly significant:

• Optical coupling: Light must be transferred efficiently between fibres and photonic chips despite large differences in scale and tight manufacturing tolerances. Small alignment errors can significantly impact coupling efficiency and system performance.

• Electrical-Optical interaction: Photonic devices are inseparable from the electronic circuits that drive them. Signal integrity, routing structures, loss, crosstalk, and latency all influence optical link performance.

• Thermal management: High-performance AI accelerators and switching ASICs generate substantial heat. Temperature directly affects wavelength stability, insertion loss, and photonic circuit behaviour, making thermal effects central to CPO design.

• Mechanical stress and packaging: Advanced packaging introduces mechanical strain and thermal expansion mismatch that can affect optical properties, create alignment errors, and reduce overall reliability.

Figure 2: Grating coupler with microlens, illustrating the field exchange between nearfield and free-space propagations.

Why traditional simulation flows fall short for CPO?
Historically, photonic, electrical, thermal, and mechanical simulations have been performed independently. A photonic designer may optimise an optical modulator without considering package temperatures, while a packaging engineer may evaluate mechanical stress without understanding its impact on optical coupling efficiency.

While this siloed approach may be adequate for discrete optical modules, it fails in CPO environments where the interactions between physics domains directly determine system performance. Designers must simultaneously address nanometre-scale photonic structures, millimetre-scale optical interfaces, and centimetre-scale package assemblies while accounting for electrical, optical, thermal and mechanical effects.

As a result, modern CPO development increasingly relies on comprehensive simulation environments that unify optical, electronic, thermal, and packaging analysis within a common workflow. Rather than treating these domains as separate tasks, they become part of a unified design workflow. For instance, Synopsys’ CPO workflows identify thermal effects, stress effects, voltage drop, electromigration, cross-die electromagnetics, and optical I/O design as tightly interconnected challenges requiring integrated multiphysics solutions.

Figure 3: Full-wave EM simulation with Ansys HFSS and signal integrity analysis including packaging effects with Ansys Circuit and Ansys Lumerical INTERCONNECT^TM.

Optical I/O: The critical interface between silicon and fibre
For many CPO systems, the greatest practical challenge lies at the optical interface itself. Even a highly optimised PIC delivers little value if light cannot be coupled efficiently between fibre and chip.

Optical I/O is fundamentally a multiscale problem. Silicon waveguides operate with dimensions measured in hundreds of nanometres, while optical fibres and packaging structures are orders of magnitude larger. Coupling technologies such as grating couplers, edge couplers, micro-lenses, and fibre attach systems must bridge these scales while maintaining high efficiency and manufacturability.

Synopsys combines Ansys Lumerical electromagnetic simulation with Ansys Zemax OpticStudio® optical system design software to address this challenge. Near-field optical behaviour can be modelled using rigorous photonic simulation, while free-space propagation and optical assembly effects are evaluated using optical system modelling. This enables designers to analyse complete fibre-to-chip coupling systems rather than isolated elements.

By combining electromagnetic simulation with optical system modelling, engineers can evaluate complete fibre-to-chip coupling systems and assess the impact of alignment tolerances, lens positioning, fabrication variation, and packaging constraints. These capabilities are increasingly important as AI systems require higher optical-port density and tighter manufacturing tolerances.

Electrical-Optical co-design
One of the defining characteristics of CPO is the close relationship between optical and electrical domains. Photonic devices do not operate in isolation; they are driven by electronic circuits, interconnected through advanced packaging and evaluated according to overall link performance.

Signal integrity directly influences photonic performance. Package routing, channel loss, crosstalk, and electromagnetic behaviour all influence overall link quality. Consequently, accurate simulation requires simultaneous consideration of electrical and optical behaviour.

Synopsys addresses this challenge through co-simulation capabilities linking photonic circuit modelling with electromagnetic analysis tools. High-frequency electrical structures can be characterised using full-wave EM simulation, while photonic circuits are represented through compact models and system-level simulation. The combined workflow enables end-to-end, electrical-optical-electrical evaluation of complete communication channels.

This integrated perspective is particularly important for AI clusters where aggregate bandwidth reaches unprecedented scales. Optimising only the photonics or only the electronics is no longer sufficient. True system performance depends on the interaction between both domains.

Figure 4: End-to-end Synopsys photonics solutions, with OptoCompiler serving as the central cockpit for component simulation, circuit simulation, and layout design.

Designing and verifying photonic integrated circuits
Every CPO system begins with individual photonic components. Waveguides, grating and edge couplers, modulators, photodetectors, ring resonators, and multiplexers must first be designed and validated before they can be assembled into larger circuits.

Ansys Lumerical FDTD™, Ansys Lumerical MODE™, and Ansys Lumerical Multiphysics™ photonic-component simulation tools now part of the Synopsys portfolio provide the foundation for this process through rigorous 3D electromagnetic and multiphysics simulation. Component geometries, process information, and foundry PDK data can be used to analyse optical performance, fabrication sensitivity, and physical behaviour at the device level. Designers can evaluate not only optical characteristics but also the impact of thermal and mechanical effects on device operation.

Once validated, the behaviour of these devices can be translated into compact models using Ansys Lumerical CML Compiler™ development software. Compact models are essential because a realistic AI interconnect may contain hundreds or thousands of photonic devices. Running full-field simulation across an entire system would be impractical. Compact models capture the relevant device physics while providing the speed required for system-level analysis.

The result is a scalable workflow in which physical device accuracy is preserved while enabling efficient circuit design. This bridge between rigorous multiphysics simulation and circuit abstraction is one of the key enablers for industrial-scale silicon photonics deployment.

Synopsys OptoCompiler further extends this workflow into an end-to-end Synopsys photonics solution, spanning component simulation, circuit simulation, and layout design.

Figure 5:Thermal analysis of a four channel WDM circuit with Ansys Redhawk

Thermal-aware and layout-aware design
Thermal management is one of the most significant concerns in co-packaged optics. As optical and electronic dies move closer together, temperature-induced performance variations become increasingly difficult to ignore.

Temperature affects wavelength stability, insertion loss, and overall circuit behaviour. In particular, ring resonators may experience resonance shifts due to thermal crosstalk. As optical and electronic dies move closer together, package-level thermal effects become increasingly difficult to ignore.

Synopsys supports a hierarchical thermal workflow spanning device, chip, and package levels. Chip Thermal Models (CTMs) enable efficient package simulation while preserving sufficient accuracy for photonic analysis. Temperature maps generated during package simulation can then be imported into photonic design environments for thermal-aware circuit verification.

Mechanical effects must also be considered. Temperature gradients create thermal expansion mismatch, inducing strain within PICs and packaging structures. This strain can modify optical properties, create birefringence, reduce coupling efficiency and generate alignment errors. Synopsys workflows incorporate structural analysis alongside photonic and thermal simulation to address these challenges. Collaboration with major foundries has further enabled integrated packaging-stress workflows for advanced 3DIC architectures.

AI-assisted design optimisation
As CPO systems become more complex, traditional optimisation approaches struggle with the sheer number of design variables involved. Fibre position, coupler geometry, package dimensions, thermal parameters, and electrical configurations can create enormous design spaces.

AI-assisted simulation is emerging as a powerful solution. Synopsys workflows use machine learning techniques to accelerate exploration of large design spaces while maintaining correlation with physics-based simulation. This enables engineers to evaluate more design alternatives than would be possible with direct simulation alone.

In demonstrated optical I/O workflows, hundreds of physical simulations were used to train surrogate AI models capable of predicting performance with high accuracy. These models enabled optimisation processes more than 500 times faster than direct simulation while maintaining close agreement with rigorous solver results.²

Figure 6: AI-assisted optimization of a lensed edge coupler using Ansys optiSLang.

The same philosophy is extending to broader system-level applications. AI-enabled reduced-order models can accelerate electrical-optical link analysis, helping engineers evaluate complex communication systems more efficiently than traditional approaches.

As AI drives unprecedented demand for bandwidth and energy efficiency, the irony is striking: AI is now helping design the optical infrastructure that makes large-scale AI possible. By combining rigorous physics with intelligent optimisation, engineers can evaluate more concepts, improve robustness and reach optimal solutions faster than ever before.

Conclusion
Co-packaged optics represents a fundamental shift in data centre architecture, bringing photonics directly into the heart of AI infrastructure. Yet the promise of CPO can only be realised through a holistic engineering approach that unifies photonics, electronics, and advanced packaging.

As AI systems continue to scale, multiphysics simulation is evolving from a design-support function into a strategic enabler. By providing a unified environment for modelling optical, electrical, thermal, and mechanical interactions, integrated simulation workflows allow engineers to accelerate innovation, reduce risk, and bring next-generation AI interconnect technologies to market more efficiently.