'Virtuoso Meets Maxwell' is a blog series aimed at exploring the capabilities and potential of Virtuoso® RF and Virtuoso MultiTech. So, how does Virtuoso meet Maxwell? Now, Virtuoso platform supports RF designs, and the RF designers measure the physical and radiation effects by using the Maxwell's equations. In addition to providing insights into the useful software enhancements, this series broadcasts the voices of different bloggers and experts about their knowledge and experience of various tools in the Virtuoso IC-Packaging world along with the nuances of RF, microwave, and high frequency designs. Watch out for our posts on Mondays.

Hello Everyone, I hope some of you have been enjoying my miniseries titled Help with Electromagnetic Analysis where I introduced the different steps involved in running Electromagnetic (EM) analysis from Virtuoso RF Solution. In Part I, I described some of the details of setting up the process stackup and in Part II, discussed how to select geometries for running an EM analysis. Today, let’s delve a little deeper in geometry selection and focus on a specific topic that deals with automatic shape preprocessing.

What’s automatic shape preprocessing? Now, let’s see, how to best describe this. Imagine that you are analyzing an inductor. Typically, in the middle of the inductor there will be an empty space. But because of metal density rules associated with manufacturing processes, the empty space will have to be filled with metal fills. A similar inductor with metal fill is shown in the picture below. This metal fill may or may not have an appreciable effect on the inductor Q, but since the fill metal shapes are very small, including these in the EM analysis will definitely impact the number of unknowns and the run time of the EM analysis. Hence, designers often want to run this inductor structure with or without the metal fill. But, “cleaning” the inductor cell will involve flattening the cell and manually removing the fill shapes, which forces the user to have multiple copies of the same layout.

Another example of layout shapes where shape preprocessing can be helpful is a shield structure underneath or around an inductor or a mesh-like structure often seen in RF ICs, where the mesh-like structure serves as the return path. An inductor coil with a similar mesh-like shield under the coil is shown in the picture below (I am also showing just the inductor coil with the shield layers turned off and just a portion of the shield structure).

The mesh-like structure is constructed using alternate layers of metal1 and 2, where metal1 goes in the vertical direction, metal2 goes in the horizontal direction, and the two are joined with via1. This inductor alone has nearly 170K shapes. In a real layout, a mesh-like structure like this can easily contain 5-6 million shapes. Modifying or simplifying structures like this by hand and then attempting multiple simplification recipes to find the optimum recipe is not a simple matter.

In Virtuoso RF solution, we give you the ability to automatically preprocess these shapes based on some recipe which can be stored and reused.

Why can’t I simply delete all of the metal fills or manually simplify the shapes before setting up my EM Model? You can definitely delete the metal fill shapes or manually simplify any mesh-like structure, however, simplifying 100s of thousands (and maybe even millions) of shapes manually is not a trivial matter. But, more importantly, it is not realistic to create different variants of preprocessing by manually simplifying the layout.

Why do I need to create different variants of preprocessed layouts? Deciding on the optimum simplification which does not change the physics and current paths of the problem but reduces the number of unknowns for the EMAG analysis is tricky and may not be obvious. Hence, having the ability to quickly and automatically generate several variants is extremely beneficial. This will become clearer as we go through the examples below.

OK, so how does Virtuoso RF solution help us for shape preprocessing? Virtuoso RF solution has built-in routines that can preprocess the shapes in the EM model before sending these shapes to the EM analysis engine. For example, going back to the inductor with floating dummy metal fills in the center of the inductor, I can set a tolerance limit so that all floating shapes will be removed from the inductor before it goes to the EM analysis engine.

The following picture shows a preview window of the effect of the preprocessing. Here, we can see that the dummy metal fills are gone. We can see also in the Layers palette to the left that there are no shapes on metal1, metal2, and metal3, thus confirming that the fill shapes from these three layers have indeed been removed. 

All of this preprocessing is done in memory and passed over to the EM analysis engine. As a result, the original layout does not have to be modified at all. So, to summarize, in Virtuoso RF solution, the shape simplification is automatic, repetitive, and extremely fast compared to any manual attempt.

The other shape preprocessing that can be done is to cluster the vias. This is done without compensating the via resistance, because if the effect of the via resistance is not accounted for correctly, then the inductor Q will be calculated incorrectly.

In the picture below, a portion of the mesh for an inductor is shown near the metal11 metal10 cross-over. You can see that the vias in the via10 layer have not been clustered. As a result, a lot more mesh elements are seen on the left-hand side 2D view; the right-hand side 3D view also illustrates the individual vias.

On the other hand, when via clustering is enabled for the same via layer, you can see below that the via appears to be clustered both in 2D and 3D views, and as a result the number of mesh elements drastically reduce at the location of the cross-over.

Let’s look at the example of the other inductor with a complex shield structure. The original shield has vertical metal1 lines and horizontal metal2 lines, and these lines are 0.2um wide and spaced 0.1um apart. The via1 vias are inserted between the metal2 and metal1 lines. I can use a preprocessing setting to first merge all the shapes in metal1 based on a minimum distance. Similarly, I can merge all shapes on metal2 and via1 layers. In the same recipe, after the merging, I can also stripe these shapes with a width of 5um and spacing of 5um. By doing this, I have effectively “simplified” the mesh-like shield structure, but have not changed the fundamental current path. The picture below on the left shows the effect of this preprocessing. The picture on the right shows the effect of a similar preprocessing, except the width is 15um and the spacing of the metal on metal1 and metal2 are set to 15um. Note that for both of the pictures, I turned off the visibility for via1 layer. The number of mesh elements (or, unknowns) for the structure on the left is 29,299 versus the number of unknowns for the structure on the right is 9256, which tells us the impact of the shape preprocessing. Run time for the structure on the left is 14m 21s, whereas for the one on the right is 3m 23s.

In case you’re curious about the impact of the preprocessing on the s-parameters, well, I was curious too. Hence, I extracted the s-parameters using AXIEM for the above two structures and calculated Q and L from them and you can see these plotted in the graph below.

In this graph, you actually see 3 sets of plots. The red lines are single ended L and Q plotted for the structure where preprocessing was used to change the shield to have 15um width and 15um spacing, the green lines are Q and L for the shield structure with 5um width and 5um spacing, and finally, the cyan lines are for a third preprocessed structure with 5um width and 2.5um spacing. As you can see, the cyan and the green lines are converging. Hence, by playing with the preprocessing recipe, I can find an optimum recipe, which can be used in the future for similar structures/layouts and the optimum recipe will still consume significantly less number of unknowns than the original shield structure with 0.2um width and 0.1um spacing.

Other simplification techniques: Besides removing floating shapes, merging metal shapes, and striping metal shapes and vias, Virtuoso RF solution also allows you to smooth the shapes. This is especially useful when there are sawtooth edges in a layout, which can also give rise to a lot of unnecessary unknowns.

So, now you know everything that’s to be known about shape preprocessing. And, don’t forget the best part. These preprocessing routines are extremely fast (gets done in a few seconds to a few minutes depending on how complex the original layout is) and is completely repeatable because the user can save the recipes with the EM model.

That’s all for today. In the next part, I will discuss about port setup and boundary conditions. That will be another important topic for you to grasp. Until then, like Tigger, I am going to say TTFN. Take care folks.

Related Resources

• Virtuoso RF Solution
• What’s New in Virtuoso (ICADVM18.1 Only)
• Video - Running an Electromagnetic Analysis Using AXIEM
• Virtuoso Electromagnetic Solver User Guide

For more information on Cadence circuit design products and services, visit www.cadence.com.

Contact Us

For any questions, general feedback, or even if you want to suggest a future blog topic, write to custom_ic_blogs@cadence.com.

About Virtuoso Meets Maxwell

Virtuoso Meets Maxwell series includes posts about the next-generation die, package, and board design flow with a focus on reinventing and optimizing the design process to ensure that the designer remains a designer! Keep watching!

Click Subscribe to visit the Subscription box at the top of the page in which you can submit your email address to receive notifications about our latest Virtuoso Meets Maxwell posts.


Sutirtha Kabir