Startup Slashes Component Design Times
Simulant's built a software program that it says models a range of complicated processes related to optical component design -- such as imitating the flow of light through a chip. The startup's founders hail from the California Institute of Technology (Caltech). They hit on their ideas while trying to create photonic crystal waveguides in the nanofabrication group at Caltech's optical lab. They say that making these waveguides usually calls for lots of costly and frustrating trial-and-error fabrication. By doing simulations in advance, they claim, it is possible to get the fabrication right the first time.
The Simulant team is using a version of the Finite-Difference Time Domain (FDTD) method to design their photonic crystal waveguides. FDTD is based on Maxwell's equations, which describe electromagnetic waves including radio-frequency, microwave, and infrared. Their method takes a guess at the answer, then repeats the same calculation over and over, and gradually homes in on the exact solution. The more complex the first guess and the longer the calculation goes on, the better the accuracy. The problem was, the team found they would need tons of processing power to crunch the numbers involved.
"We didn't have millions to spend on a Cray [supercomputer], so we decided to create our own clustering software," says cofounder Michael Hochberg. Clustering software is a program that allows ordinary PCs to be connected together to perform computational tasks that usually call for a supercomputer. Hochberg and another student, Tom Baehr-Jones, worked on the project with their professor, Axel Scherer, a leader in his field.
The results got the job done, and then some. "We spent around $80,000 for 32 machines and wound up beating a Cray T3D in terms of performance and memory," Hochberg says. The team figures it reduced the overall development time to between one-fiftieth and one-hundredth what it would have taken with traditional methods and machinery.
Simulant says its patented clustering software, called the Feonor System, works by putting an application through an algorithm that enables it to be distributed across multiple computers, PCs or otherwise, which in turn are networked (Simulant is using fast Ethernet). Once distributed, the software assigns specific functions in the application to processors in the cluster, calling in more processors as needed and generally load-balancing across the computers to get the job done efficiently.
Simulant says they haven't yet hit a limit on the number of PCs that can be clustered with the Feonor System. The biggest application, they say, is running on 50 dual-processor PCs in the network of their first beta customer, an unnamed optical component startup in California.
Photos of Simulant's founders on their Website look like they belong in a high school yearbook. But make no mistake, these students are launching a very grownup bid for their share of one of the world's hottest markets. The startup's just won $400,000 in seed funding from Information Technology University Ventures, a firm that specializes in launching technology startups from teams on college campuses. ITU's other optical venture, Opient, offering a platform for large-scale integration of optical components, was also launched in July (see Optical Investors Go to College).
Simulant plans to launch products, hire help, and go for more funding within a couple of months. So far, in addition to Hochberg, Baehr-Jones, and Scherer, the management team includes COO Adam Lawton, also a Caltech undergrad, and an unnamed senior software engineer.
Whether these initial steps lead to more funding and greater horizons remains to be seen. But signs are good, say the founders. In addition to its beta customer (which Hochberg says is not Opient), the team says it has a trial coming up with a "Fortune 500" optical component supplier.
-- Mary Jander, senior editor, and Pauline Rigby, senior editor, Light Reading http://www.lightreading.com