Yesterday, in my post System Analysis: Computational Software at Scale, I talked about computational software at scale and the three aspects that Cadence is delivering:

• Algorithms that operate on unimaginably huge amounts of data
• Scaling algorithms to be massively parallel in the cloud
• Delivering and supporting software in an enterprise environment

Cadence has delivered this approach to technology in a number of products that scale to large numbers of processors. In the simulation world, there are the Xcelium and Spectre X products. At the system level, there is the Clarity 3D Solver and the Celsius Thermal Solver. I'm not giving away any secrets when I say there will be more.

For details of these products, see what I wrote about them when they were announced:

• What Is RocketSim? Why Did Cadence Acquire Rocketick?
• Spectre X: Same Accuracy, New Speed
• Bringing Clarity to System Analysis
• Celsius: Thermal and Electrical Analysis Together at Last

Clarity 3D Solver

At the recent DesignCon, Bruce Yuan gave detailed presentations on the Clarity 3D Solver in the context of More than Moore, or as it was described accurately on the title slide, 3D Full-Wave Simulator for Advanced Packages. His first slide showed how Cadence is extending its algorithmic know-how from chips to system analysis, as I discussed above:

At the bottom are the existing design platforms in our portfolio: Virtuoso and Innovus for chips, SiP and OrbitIO for complex packaging, Allegro and OrCAD for PCB, and our partners Dassault and PTC for chassis. In the middle are the multiphysics domains: EM, electrical, thermal, fluid-flow, mechanical. And, at the top, the various engines that partition and do the actual solving. Together, these allow us to do system design optimization, a key part of what we call Intelligent System Design.

As I said, under the hood there is a massively parallelized matrix solver. This is a breakthrough algorithm and is part of Cadence's secret sauce in the system analysis area. It has near-linear scalability without any accuracy loss. It also has virtually unlimited capacity using a large number of low-capacity machines without requiring any huge machines that are either unavailable when you want them, or are sitting idle much of the time waiting for you to show up to use them. The whole infrastructure is dynamically deployed into the cloud (or a data center) and has fault-tolerant restart, since, with huge numbers of machines, rare things happen regularly.

Bruce showed the legacy flow, which does the adaptive meshing on a single master. In practice, on a big design, this might require a multi-terabyte machine.

Instead, our system analysis tools, and Clarity 3D Solver in particular, distributes the meshing, too, as well as distributing all the different frequencies. 

Results

Here are a couple of results:

• A wafer-level SiP for HBM. This is a 4-layer RDL wafer-level SiP structure with two chips, an SoC and an HBM memory bare die. Signals are 2um width and spacing, 128 full-channel data signal plus power delivery. Full analysis with Clarity 3D Solver on 128 cores took 7.2 hours (compared with over 86 hours with legacy). The memory required was 22GB, a reduction of 85%.
• A die-stacking SiP. This is a 4-layer BGA substrate with three chips, an SoC die, and 2 DRAM die, with an RDL included in the top DRAM die (see picture). Analysis of the RDL and SiP on 15 cores with Clarity 3D Solver took 3.5 hours (compared to over 26 with legacy). The memory required was 20GB, a reduction of 87%.
• A third example, for customer Socionext, just looking at the cost in the cloud using Cadence Cloudburst on Amazon AWS (EC2 instances). Memory is 256GB, with 64 cores, at $3.20 per hour. Total runtime was 60 hours for a cost of $192. The legacy approach needs 1TB of memory, 64 cores at $6.67/hour (due to increased memory). Total runtime was 440 hours for $2935. Clarity 3D Solver is 93% cheaper to run.

Celsius Thermal Solver

The other system analysis tool that Cadence announced last year was the Celsius Thermal Solver. At DesignCon, CT Kao presented an overview of the product.

This performs simultaneous thermal and electrical analysis for ICs, packages, boards, and systems. Under the hood, yes, finite element analysis (FEA), so more simultaneous solutions of large sets of equations, so more sparse matrices. In addition, there is computational fluid dynamics (CFD) to handle airflow and fans in cooling systems, but that is also more of the same.

For the IC-centric approach, Celsius Thermal Solver doesn't just address the obvious stuff but also 3D-IC, die-to-die bonding, and thru-silicon-vias (TSVs), delivering a temperature and power map across the die taking all those complexities into account when present.

For the package/PCB-centric approach, Celsius Thermal Solver uses FEA and CFD for both transient and steady-state simulation. It resolves detailed 2D layered and 3D structures.

For the system-centric approach, in addition to the FEA and CFD techniques, Celsius Thermal Solver can also handle heat-sinks, enclosures, airflow, and all the other aspects of the system environment that affect temperature.

Results

I'll just give one example, which I believe is a camera constructed on three small circuit boards, which then goes into some sort of enclosure. Anyway, as you can see from the image, it is something like that.

There are almost 100M elements that make up the analysis. By scaling to a large number of cores, the results are produced over ten times faster. In some cases, Celsius Thermal Solver enables analysis of large systems that couldn't be analyzed previously.

More Details

In addition to my blog posts linked to at the start of this post, there are also the Clarity 3D Solver product page and the Celsius Thermal Solver product page.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.