Cloud native EDA tools & pre-optimized hardware platforms
Multi-mode operation and polarization conversion/splitting are common challenges in photonics integrated devices and circuits. It can be especially difficult to model them correctly and efficiently. In this report, we will demonstrate how to use the 草榴社区 full design flow, combining device and circuit simulation tools as well as a layout tool for a photonic integrated circuit (PIC).
To demonstrate the capability of our simulation software, we select a relatively simple PIC as shown below[1] in Figure 1:
We picked this example because we would like to validate our circuit tool by comparing it with direct device simulation of the whole structure. This is the largest structure a device simulation tool can handle on a regular PC even at very coarse mesh.
There are two approaches:
1. Direct device simulation of the whole structure
2. Circuit simulation with custom PDKs by decomposing the whole structure into three sub-devices: polarization rotator, coupler, and splitter.
For the device simulation, the tool we use RSoft RCWA-based ModePROP, which is relatively efficient but maintains accuracy.
To simulate the whole structure using a device tool, a very coarse mesh must be used (Dx=Dy=40nm, Dz=1?m) in order to run it on a regular PC with 64G RAM. Shown below in Figure 2 is the layout of the whole structure in RSoft CAD, with an aspect ratio of 20:1.
Figure 2. Layout of the whole structure in RSoft CAD
TE0 and TM0 modes, as shown below in Figure 3 , of the input 450nm wide waveguide are dynamically calculated for input launch field as well as output overlap.
(a) TE0 mode
(b) TM0 mode
Figure 3. Mode profiles of the input waveguide with 450nm width
The modes of the two output waveguides are also dynamically calculated as launch field and overlapping mode, shown below in Figure 4.
(a) TE0 mode of the straight output waveguide
(b) TE0 mode of the bent output waveguide
Figure 4. Mode profiles of the output waveguide with 500nm width
At a very coarse grid (b), one simulation (one input and one wavelength) takes about 40G RAM and 43 minutes on a 16-core PC. Shown below in Figure 5 are the simulation results for TE and TM inputs.
Figure 5. Device performance by device simulation
Next we’ll show the field patterns, with an aspect ratio of 20: 1, for both inputs (upper TE, lower TM) at wavelength 1.55?m.
Figure 6. Field patterns for input TE (upper) and TM (lower) modes
As mentioned, the whole structure is decomposed into three sub-devices: polarization rotator, coupler, and splitter. This layout is displayed in the RSoft CAD in Figure 7.
Figure 7. Decomposition of the whole device into three sub-devices
The RSoft built-in Custom PDK Utility can be used to automatically generate a PDK. A PDK includes an S-matrix for circuit simulation, and geometric framework for layout, for each of the sub-devices.
The polarization rotator shown in Figure 7(a) is a device with 1x1 physical ports. Each port carries two modes so it is 2x2 virtual ports from a circuit point of view. The input modes are the same as shown in Figure 3 used for the whole device simulation. The output modes are multi-modes guided in the wide (850nm) waveguide, as shown below in Figure 8.
(a) TE0 mode
(b) TE1 mode
Figure 8. Mode profiles of the input waveguide with 850nm width
Shown below, S-matrix for input TE0 (Port 1) and TM0 (Port 2) modes:
Figure 9. S-matrix of the polarization rotator
It is observed that both cross-talk and backward reflection is very small.
The coupler shown in Figure 7(b) is a 2x2 port device both physically and virtually. The input modes are the same as shown in Figure 8. The output modes are each TE0 mode from the two physical waveguides with different widths, as shown below in Figure 10.
(a) TE0 mode with 650nm width
(b) TE0 mode with 500nm width
Figure 10. Mode profiles of the output waveguides
Figure 11. S-matrix of the coupler
Some cross-talk is observed, as well as some backward reflection because of appearance of the 2nd waveguide.
For the splitter shown in Figure 7(c), it is also a 2x2 port device both physically and virtually. The input modes are the same as output modes of the coupler, shown in Figure 10. The output modes are each TE0 mode from the two physical waveguides with different width, as shown below in Figure 10. The output modes are the same as the output modes of the whole device shown in Figure 4.
Figure 12. S-matrix of the splitter
With the established custom PDKs, we can build a compound component (CC) in RSoft OptSim Circuit, a schematic layout for the whole device, as shown below.
Figure 13. Schematic layout in OptSim Circuit
With the full integration of RSoft simulation tools and the PIC Design Suite, the device can be physically laid out in OptoDesigner automatically, as shown below.
Figure 14. Physical layout in OptoDesigner
With an aspect ratio of 1, it is a very long and narrow device, as expected.
With the S-matrix data for each sub-device, we can test the circuit performance in OptSim Circuit with a proper input source (wide-band) and monitors (Spectrum Analyzers), as shown below.
Figure 15. Test setup in OptSim Circuit for performance simulation
Shown below, the circuit performance of the PIC.
Figure 16. Performance simulation results in OptSim Circuit
Compared with the direct device simulation results shown in Figure 5, circuit simulation gives very similar results. Detail comparison can be summarized in the following table, for 1.55?m wavelength, only.
|
TE0 input |
TM0 input |
||
|
Device simulation |
Circuit simulation |
Device simulation |
Circuit simulation |
Transmission |
0.300200 |
0.3016643 |
0.238606 |
0.244830 |
Cross-talk |
0.000468 |
0.0004202 |
0.000735 |
0.003042 |
Back-Reflection |
0.000001 |
0.0000098 |
0.000076 |
0.000167 |
Cross-Reflection |
0.000012 |
0.0000065 |
0.000012 |
0.000028 |
Overall, device and circuit simulation results agree well. Some discrepancy, however, is observed, especially for TM0 input. One reason is that two waveguides are separated by 0.2?m only, as well as some mode coupling in the initial stage of the splitter, as shown below in Figure 17.
Figure 17. Mode coupling in splitter
More accurate circuit simulation results can be obtained by combining coupler and splitter together as one PDK, at the cost of much longer computation time and more computer memory
Shown below are new circuit results with combined coupler and splitter as one PDK.
|
TE0 input |
TM0 input |
||
|
Device simulation |
Circuit simulation |
Device simulation |
Circuit simulation |
Transmission |
0.300200 |
0.2995084 |
0.238606 |
0.237913 |
Cross-talk |
0.000468 |
0.0012398 |
0.000735 |
0.001324 |
Back-Reflection |
0.000001 |
0.0000011 |
0.000076 |
0.000293 |
Cross-Reflection |
0.000012 |
0.0000001 |
0.000012 |
0.000001 |
As observed, the circuit simulation results are closer to device simulation results.
Another way to improve the accuracy is to count all the possible modes inside, such as TM0 and TM1 mode of the polarization rotator. They can be coupled back into TE modes at the end, more or less. The computation effort, however, could be doubled at least. This could leave for future investigation.
It has been demonstrated that multi-mode and polarization dependence/coupling can be handled correctly in RSoft’s Photonic Device and circuit system tools. The simulated PIC can be exported to
草榴社区 OptoDesigner directly for layout. The whole PIC design flow is fully automated and seamlessly connected.
Please be advised that the purpose of this example is to show the consistency between device and circuit simulation tools, not to make a comparison with the published results, since very coarse meshes are used in the device simulation. Otherwise, it is not feasible to perform the device simulation for the whole structure and compare it with the circuit simulation.
Simulation files are available upon request by emailing photonics_support@synopsys.com.