Applications for Acceleration

Since its inception, Machine Discovery’s goal has been the acceleration of complex simulations. And at a basic level, the value of running simulations quickly seems self-evident: It saves time, it saves resources, it reduces time-to-market. These advantages, though, don’t fully reflect the impact of what acceleration provides. Yes, in an incremental sense, the ability to do the same thing faster is enough to shift the bottom line — often substantially. With a bit of creativity in application, though, acceleration opens up much more: New ways of working, new approaches to design, solutions to problems that were previously intractable.

Our product team’s focus, then, has been to empower users to exploit our technology in new ways — and, in doing so, maximise the value they receive from it. In support of this aim, we’ve been expanding our Discovery Platform’s API. An API — an Application Programming Interface — essentially provides a way for computer programs to talk to each other. In our case, we’ve enabled users to access the Discovery Platform’s optimisation functionality, statistical sampling and emulator predictions directly from their own programs and scripts.

As one example, our new API enables users to query their trained emulators rapidly and efficiently across a broad range of parameters — gathering tens of thousands of samples within minutes — and use the results for more in-depth statistical analysis of their models. The Discovery Platform’s Graphical User Interface continues to provide easy and fast point-by-point inspection; the new API then complements this with the capabilities required for detailed exploration.

This is still fairly abstract — so let’s focus on a concrete case. Suppose we have a circuit design that we want to thoroughly test — i.e. we want to find conditions under which the circuit stops behaving as expected. The Discovery Platform and our emulator technology allow us to do this much more efficiently and effectively than via running simulations alone:

• Upload some sample data (combining existing test data and new simulation outputs) to the Discovery Platform
• Create an emulator for our design with a few clicks
• Sample the parameter space exhaustively using the Discovery Platform API (in minutes, rather than days/weeks)
• Weight our samples according to their behaviour to visually identify potential problem areas
• Focus on the problem areas — either passing on the details to customers or re-designing/re-testing with these areas in mind

Not only is this orders of magnitude faster than directly running simulations for a wide parameter range, it improves coverage and dramatically reduces the chance of missing faults. The result is a reduced time-to-market with a far lower risk of customers running into issues — so not only faster delivery, but also a higher quality, more reliable end-product.

The enhancements we’re making are already helping to address a host of new challenges like this one — from investigating out-of-spec behaviour to Monte Carlo analysis. More than this, though, the API is giving users the power and flexibility to discover their own new applications for our technology.

If you have an application in mind for your own work, let us know.