Get email delivery of the Cadence blog featured here
Despite CurvyCore sounding like something that you might take classes in at your local gym, it is actually new technology that allows computing and representing non-Manhattan shapes in Virtuoso.
The CurvyCore technology is targeted at a wide range of applications such as microfluidics, MEMS, and conformal routing. But the initial application driving the technology is photonics. Photonics involves many strange shapes, far from Manhattan geometry—let's say Lombard Street geometry, after the crookedest street in the world (actually, Lombard Street isn't even the crookedest street in San Francisco, that would be Vermont Street on the back of Potrero Hill, but that's not a tourist area).
I sat down with Gilles Lamant, the Cadence Distinguished Engineer who is leading the CurvyCore development. For a basic introduction to silicon photonics, see my post Silicon Photonics covering a presentation/tutorial that he gave to Cadence's engineering organization.
The picture above gives you an idea of just how "non-Manhattan" we are talking about. This is a very different direction in terms of layout from where advanced processes have been heading. With technologies like self-aligned double patterning (SADP), advanced processes are even more regular than just Manhattan layout, sometimes with only one dimension allowed on a layer, sometimes with only certain widths and spacings permitted. However, having both of these in Virtuoso means that both the ultra-regular FinFET geometry and the curvy geometries are supported in the same tool at the same time, as you can see at the bottom of the picture above where the rectilinear electrical connections are mixed with the curvy optical ones.
Unlike regular IC layout, you can't really draw layout like this by hand with adequate accuracy. Instead, at the core, is a mathematical model so that we can compute paths, offsets, ribbons, boundaries, and do mathematical operations equivalent to Boolean operations. The table above shows the data model, which consists of three main layers. The pink layer at the top consists of the actual polygons of the layout, OpenAccess shapes. Since OpenAccess doesn't support curves, there can be a large number of polygons to represent one of these layouts with adequate precision. The middle blue layer is an IEEE double floating point model, with discretized shapes. The bottom green layer is a purely mathematical model, interfacing through symbolic equations. In effect, the green at the bottom is an accurate mathematical representation of the shapes, the top pink layer is "polygon level layout", and the intermediate layer serves as a sort of translation layer between mathematics and polygons, preserving as much precision as possible without going all the way to arbitrary precision arithmetic.
For the rest of this post, I'm going to focus on photonics. One reason for this is that Cadence is holding a photonics workshop on November 7th and 8th. See the end of this post for details.
For the last few years, photonics has largely been driven by datacenter and HPC requirements. Since datacenter networking is mostly optical, they have moved heavily into photonics. But that has opened the door to other applications, too, such as Lidar, biomedical, and military. 5G is starting to drive RF directly over fiber (without first digitizing it). Some people see photonics as the replacement for the current SerDes approach, which is starting to require too much energy. From a business point of view, foundries have had to bring up photonics capability to satisfy the HPC market, but having developed the technology these other markets are opportunities.
The reason that CurvyCore plays nicely with photonics is that light is not like electric current. You've probably heard that light likes to travel in a straight line, and that is true. But light really doesn't like to go around sharp corners since they reflect and cause signal losses—the same effects happen with electrical signals once you get up to RF frequencies. To make light go around a corner you have to do it gradually around a curve, hence CurvyCore. Light is unlike electrical signals, even RF, since it has different modes (different colors aka frequencies, different polarizations, electric and magnetic field components).
An attractive application area is in planes and satellites since optics is radiation-robust. In satellites, there are particles coming from the sun, but photonics is well-behaved under those conditions. And it is way lighter. It is unclear where its role will be in automotive since that is extremely cost sensitive (the other end of the scale from satellites), although anything that reduces weight is attractive.
Another attractive area is the microfluidic medical market. This is a mixture of photonics (for optical sensing) and microfluidics, involving channels for moving almost infinitesimal amounts of samples and reagents, and taking measurements. They are very complex, involving a lot of computation so requiring advanced processes on the same die. It has the potential to be a big market since these devices are use-once-and-throw-away. Like an inkjet printer cartridge, each one contains a chip, but if you are monitoring your blood in a home lab, you will get through many more units than you do printer cartridges.
As I said above, Cadence is holding a photonics workshop on November 7th and 8th. The first day is a series of presentations from people working in photonics. The keynote is by Vladimir Stojanovic of UC Berkeley titled More than Moore with Electronic-Photonic Integration.
The rest of the day includes:
The second day is a hands-on photonics workshop with Lumerical and Mathworks. The layout part will all be done using the CurvyCore version of Virtuoso, so you can get to play with it directly. The example task used in the workshop is designing a Lidar photonics IC:
More details, including a link for registration, are on the Photonics Summit Page.
Sign up for Sunday Brunch, the weekly Breakfast Bytes email.