Photonic integrated circuits (PICs) are a key enabling technology for a broad range of current and next-generation products. Combining semiconductor materials and manufacturing processes common to microelectronics with the encoding, transmission and detection of light, PICs are transforming communication in datacenters by bringing bandwidth closer to compute cores, and are accelerating emerging applications like LiDAR for autonomy and quantum computing for the future of information processing.
The connection between the electronic and photonic realms is made by devices capable of encoding electrical signals on optical channels and recovering that information by converting light back into electrical charge. In PICs, electro-optic modulators and photo detectors are the fundamental optoelectronic components that make these transitions possible.
Ever-increasing demands for bandwidth, power efficiency, and sensitivity require leading-edge simulation technology that connects the device model with the manufacturing process and its complete multiphysics behavior. Combining Silvaco Victory Process and Ansys Lumerical software for TCAD-enabled photonic device simulation equips designers and engineers with the necessary tools for complete and accurate prediction, analysis and optimization of optoelectronic device behavior.
Optoelectronic component design for photonic integrated circuits (PICs) starts with accurate modeling of the physical structure and the doping profile that defines the optical and electrical behavior of the device. The objective is the creation of a physical model that reflects the as-manufactured component. The design process starts with inputs from the fabrication process: materials and mask patterns combined with etch, implant, annealing and growth conditions. While a geometric CAD model of the structure can serve as a starting point for early-stage design exploration, process simulation with Silvaco Victory Process is necessary to make the connection between the manufacturing steps and the resulting physical structure. In Figure 1, the workflow for photonic device simulation using inputs from Victory Process is illustrated.
Fig 1. Ansys Lumerical photonic device simulation workflow with TCAD input from Silvaco Victory Process
Geometric effects such as etch-influenced side-wall angles and layer interfaces from conformal deposition are important for the accurate simulation of light propagation . In optoelectronic devices, the definition of the implant profile, which is constrained by the manufacturing process, is critical for achieving optimal performance trade-offs for figures of merit including modulation efficiency; dark-current and related detector sensitivity; and bandwidth. Here again, Silvaco Victory Process is necessary to connect these specific behaviors to the manufacturing inputs.
Once the physical structure, including the material interfaces and doping profile, has been simulated, it can be easily exported from Silvaco Victory Process and imported into the Ansys Lumerical simulation tools. This automated data exchange process ensures that geometry and materials are accurately mapped between software and that the most accurate representation of doping profile from the process simulation is maintained.
The results of the process simulation form the input into the next stage in the optoelectronic design workflow: device simulation. Multiphysics simulation is critical for the prediction, analysis and optimization of device performance. Based on the physical structure input, multiple aspects of the device performance can be simulated, including eigenmode analysis of waveguides, optical propagation and absorption, optical to electrical conversion, charge transport, electro-optic material response, and thermal behavior. Depending on the behavior of interest, multiple Ansys Lumerical solvers can be used to predict and analyze performance. For example, the modulation response of an electro-optic modulator can be characterized from charge transport simulation (CHARGE) and eigenmode analysis (MODE), and thermal effects can be analyzed using the HEAT solver. Similarly, the detector responsivity and bandwidth can be simulated using electromagnetic propagation (FDTD) in conjunction with CHARGE.
Process Simulation: Victory Process
The two structure types were created using one of two types of 3D process simulator, either a meshed based methodology, more suitable for large structures, or level set based, more suitable for detail oriented moving boundary simulations, such as emulating realistic etch/deposit machines, physical oxidation or stress-based deformation. If the simulated structure is not an obvious fit into either of these categories, then either simulator can be used.
The two process simulators are designed to replicate the process run sheet in a typical fabrication facility. One of the inputs is therefore a mask set, which can either be read in as a standard GDS2 format, or can be created by the user on the fly using X-Y co-ordinates in the process flow input file, which can then be converted into a viewable GDS2 mask set as the process simulation progresses.
The structure is then constructed in the same manner as occurs in a typical fabrication facility, by simply proceeding down the process run sheet using etch/deposit/implant/diffuse steps along with any relevant mask layer at each section in the process run sheet.
Each process step can use one of several models for that step, depending on the simulation detail and/or accuracy required. As with all simulators, the level of required detail and the time it takes to simulate that particular step, together with computational resources, are correlated quantities. For example, implantation can be by way of look up tables of concentration versus depth (a good option for flat surfaces), or alternatively, each individual implanted ion can be simulated separately and highly accurately as it collides with each atom in the substrate, a technique known as Monte-Carlo simulation (ideal for complicated surface topography). At each step, these optional technique choices can be "mixed and matched" depending on the importance of that particular process step, providing great flexibility to the user to prioritize and optimize accuracy versus simulation time for the simulation as a whole.
The two structure types, used by way of example in this paper, are a silicon optical waveguide with integrated germanium photodetector  and a phase shifting based optical intensity modulator using diodes and integrated transmission lines, to provide the variable electrical field as device electrical input .
Two types of structure variation were investigated. For the silicon waveguide with integrated germanium detector, the effect of top contact design on optical performance compared a large contact area top contact to the germanium detector, with a top contact using vias at the edge of the structure. An example of one variant is shown in Figure 2. In this case, the level set based process simulator was used.
Fig 2. Silicon waveguide structure fabricated on an SOI substrate, with integrated germanium photodetector showing general structure with absolute net doping as the color contour.
For the optical modulator structure, the effect of doping concentration forming the modulating diode structure was investigated using two different implant doses for the n-type and p-type implants. One structure used implant doses of 1.5e13/cm2 and 1e13/cm3 for the active region phosphorus and boron implants respectively, whilst the second experiment used implant doses of 3.2e12/cm2 and 2e12/cm2 for those same implants. The effect on dopant distribution within the optical waveguide/diode region is shown in Figure 3 where the color contours show the absolute net doping concentration for the two cases from the two different implant doses.
Fig 3. Net doping concentration for two different implant doses to investigate the effect of doping concentration on optical modulator performance.
This second example, which has very large features, such as the transmission lines, together with very small features in the integrated electric field, optical phase modulated waveguide, uses the mesh-based process simulator to reduce the computational resources required for simulation. Figure 4 and Figure 5 show the huge feature size range required to correctly simulate the structure. Figure 4 shows the complete structure which is dominated by the two transmission line metallization. In between the transmission line, just visible in Figure 4, is the integrated waveguide and integrated diode structure, which must be resolved correctly for optical and electrical characterization. A zoomed in graphic of this active diode modulator, together with striated diode doping features to reduce transmission line losses, is shown in Figure 5.
Fig 4. A transmission line based optical phase modulator structure using an integrated optical wave guide and diode structure to provide electrical field as the phase modulating mechanism.
Fig 5. A zoomed in detail of the structure shown in Figure 4, showing the striated doping pattern in the integrated diode structure to reduce transmission line losses, with absolute net doping as the color contour.
In the following sections, these process simulated structures will be imported into Ansys Lumerical software for optical and electrical simulations.
Photonic Simulation from Process Simulation Results
Photonic and optoelectronic device simulations require specific inputs to define the structure, materials, and boundary conditions for the model. Ansys Lumerical software provides a full 3D design environment and a comprehensive material library that is purpose-built for photonic device design and simulation, enabling the straightforward creation of parametric device models. Leveraging built-in interoperability technology, the automated import of process simulation results from Silvaco Victory Process seamlessly handles three key steps: structure extraction, material assignment, and doping profile definition.
Structure extraction creates 3D solid objects from finite element simulation meshes, which may contain multiple sub-domains. These solid objects are placed in the 3D CAD environment. Reconstruction of CAD geometry that is suitable for further geometric processing (e.g. Boolean operations) is challenging: a tessellated volume may have 10-100k faces or more, exceeding the capacity of most geometry kernels. Ansys Lumerical’s interoperability tools automatically identify sub-domain surfaces and simplify the extracted structure so that it can be used in the 3D CAD environment while preserving the underlying structure geometry that the mesh represents. Using this approach each volume from the Silvaco Victory Process simulation output can be imported into the Ansys Lumerical design environment while maintaining the flexibility to further adjust and modify the simulation structure, as shown in Figure 6.
Fig 6. Structure extraction from process simulation creates solid objects with associated material assignments that are suitable for further processing in a 3D CAD environment.
In addition to structure extraction, correct material assignment to each domain is critical to maintain the fidelity of the model between simulations. When extracting structures from the process simulation output, the automated import utility will also ensure that material identifiers are mapped between simulations. As a result, the correct material definitions will be associated with the geometric structures imported into the Ansys Lumerical design environment, and those material definitions will contain the necessary parameters for the physical optoelectronic simulations.
Finally, the process simulation contains information about the dopant species and the spatial distribution of the impurity densities. These are important inputs for the simulation of the device’s optoelectronic response, and maintaining the fidelity of that data is critical for achieving accurate results. Using the interoperability features in the Ansys Lumerical CHARGE solver, doping profiles can be automatically extracted and imported from the Silvaco Victory Process results and included in the charge transport simulation. Doping profiles are aligned with geometry and can be applied to specific simulation domains. The Ansys Lumerical CHARGE solver will automatically adapt its simulation mesh to conform to the spatially varying dopant densities, ensuring that the dopant profile is accurately represented in the charge transport simulation.
Having imported the structures, material domains, and doping profiles from the Silvaco Victory Process simulation into the Ansys Lumerical design environment, the physical structure of the device is now prepared for simulation. Users can further add or modify geometry, specify boundary conditions, and configure the simulation as necessary. Electrical contacts can be defined to set a DC or transient stimulus in a charge transport simulation, and optical sources can be specified to inject light into the device. Simulations can then be configured for DC, AC or transient analysis for charge transport and broadband light propagation or eigenmode analysis for photonics, enabling a comprehensive range of multiphysics analysis for these devices.
Optoelectronic Results Extraction
Photodetectors are a key component in photonic integrated circuits (PICs), enabling monolithic electro-optic systems. Using a material with strong absorption at the design wavelength, photodetectors convert optical signals into electrical signals. In silicon photonics, germanium is a common material choice since it is compatible with most silicon processes and can be grown on top of silicon with few defects. In a vertical configuration, the germanium absorber is grown on top of a silicon waveguide, and an electrical contact is made on top of the germanium. To minimize electrical loss at this contact, a thin layer of high-concentration dopants is introduced at the interface between the germanium and the contact, while the rest of the germanium is not intentionally doped. The underlying silicon is doped to increase conductivity, creating a vertical PIN junction. When the optical signal propagates along the waveguide and enters the absorption layer, the absorbed photons generate electron-hole pairs in the germanium, which are separated by the internal electric field and flow through the electrical contacts to form the output electrical current.
The vertical photodetectors (VPDs) simulated in this study use geometry and material properties from  to create a reference device. Following the proposal outlined in , we evaluate the impact of different contact geometries on the VPD performance: contact metallization is a strong absorber of photons at the germanium interface, which reduces the number photo-generated electron-hole pairs that can contribute to the current. Placement of the contact interfaces will influence dark current, responsivity, and bandwidth, and can be analyzed effectively using inputs from process simulation. Figure 7 illustrates the geometry from the Silvaco Victory Process simulation for two cases, “large” and “small” contacts, imported into Ansys Lumerical CHARGE using the workflow described in the previous section.
Fig 7. Perspective views of the 3D imported structures from Silvaco Victory Process simulator into Ansys Lumerical CHARGE solver with (a) large and (c) small electrical contacts, respectively; (b) 2D cross-sectional views of the imported structures in Ansys Lumerical CHARGE solver (b) large and (d) small electrical contacts, respectively.
Figure 8 compares the simulated electric field intensities in the two VPD configurations (“large” and “small” contacts). In the small contact case, more of the light incident on the germanium is absorbed, increasing the responsivity. More details of the simulation methodology can be found in Ref. 6.
Fig 8. 2D Lateral electrical field distributions in the simulated devices with (a) large and (b) small electrical contacts, respectively, in Ansys Lumerical FDTD
Table 1 summarizes the essential figures of merit from the simulated devices and compares the influence of large and small electrical contacts. In conclusion, the simulated device using smaller electrical contacts demonstrates improvement of 38.3% on its responsivity while maintaining low dark current and high bandwidth operation. Using process simulation in combination with photonic device simulation, the materials, structure, and doping profile of the VPD can be further optimized to improve the responsivity before fabricating devices, identifying promising designs while minimizing R&D cost.
Table 1. Comparison of key figures of merit for the vertical photodetector.
|Dark Current (nA)
Mach-Zehnder modulators (MZMs) are a popular type of electro-optic modulators used in PICs to encode electrical signal onto optical carriers. These devices employ an interferometer-type structure, balanced or unbalanced, and control the amplitude of the output optical signal via constructive or destructive interference by introducing additional phase shift on either of the arms. These additional phase shifts can be introduced by various means. For our device, we chose a depletion type MZM, which drives the PN junctions on the arms (waveguides) of the interferometer into reverse bias to deplete them of free carriers. The change in the free carrier density modifies the effective index of the waveguide through the plasma dispersion effect . Therefore, simulating such a device requires solving multiple physics.
Fig 9. (a) Perspective view of the 3D imported geometry from Silvaco Victory into Ansys CHARGE, (b) z-normal view of the imported geometry in MODE with the orange rectangle denoting the simulation region and the purple region showing the imported charge density data from electrical simulation.
For this study, we have reconstructed the design from  as a benchmark and will evaluate the impact of different doping splits on the modulator performance. Figure 9 shows the perspective view of the imported geometry from process simulation in Silvaco Victory Process. As explained in the previous section, the import process generates 3D geometry, assigns material definitions, and imports doping profiles. Once the import is complete, voltage boundary conditions are applied to the metal contacts on both sides and the voltage across the device is swept from 0.5 V to -4 V to model the electrical behavior of the PN junction under reverse bias. Figure 9 (b) also shows the XY view of the same geometry imported into the Ansys Lumerical MODE waveguide design environment. The simulation region, shown by the orange rectangle does not include the metal contacts here since they are placed far away from the waveguide core and thereby do not interact with the optical mode. The purple region shows the imported carrier density profile from the CHARGE simulation to model the perturbation to the optical model as the voltages across the metal contacts vary.
Fig 10. (a) Free carrier density in the PN junction under 4 V of reverse bias in units of cm-3, (b) small-signal capacitance from CHARGE simulation shows good agreement with measurement in Ref. , (c) additional phase shift at the end of one arm of the interferometer as a function of applied voltage, (d) optical loss on each arm as a function of applied voltage, (e) transmission spectrum from INTERCONNECT simulation showing good agreement with (f) measured spectrum reported in Ref. .
Figure 10 illustrates the main results from the simulation of depletion-mode phase shifter with nominal doping, including the transmission spectrum for the corresponding MZM. More details of the simulation methodology can be found in Ref 5.
When designing an MZM, designers optimize key performance metrics including modulation efficiency (the amount of phase shift for a given applied voltage), electro-optical bandwidth, and optical loss. The overlap between the waveguide mode and the PN junction is an important factor in device performance: a larger change in free carrier density increases modulation efficiency, but the higher carrier density also increases loss. Process simulation is a valuable tool to aid designers with evaluating this tradeoff. Figure 11 compares the modulation efficiency and loss between the two doping splits: nominal and low. By reducing the implant density (“low”), the designer creates a phase shifter suitable for a low loss modulator (Figure 11 (d) vs. (c)), but reduces the phase shift per unit length (Figure 11 (b) vs. (a)).
Fig 11. Modulation efficiency comparison between nominal and low doping splits for the depletion mode phase shifter.
- Xu Wang, et al., Opt. Lett., 39, 5519 (2014) https://doi.org/10.1364/OL.39.005519
- R. Soref, et al., IEEE Journal of Quantum Electronics, vol. 23, issue 1, pp. 123-129, January 1987. http://dx.doi.org/10.1109/JQE.1987.1073206
- Shirong Liao, et al., Optical Express, vol. 19, no. 11, pp. 10967-10972, 2011. https://doi.org/10.1364/OE.19.010967
- T. Baehr-Jones et al., Optics Express, vol. 20, no. 11, pp. 12014-12020, 2012. https://doi.org/10.1364/OE.20.012014
- Ansys, Inc., “Travelling Wave Mach-Zehnder Modulator”, ansys.com. https://optics.ansys.com/hc/en-us/articles/360042328774
- Ansys, Inc., “Vertical Photodetector”, ansys.com. https://optics.ansys.com/hc/en-us/articles/360042957893