January
15, 1998
The economics of modeling
Wolfram Blume
The gap between the high cost of developing models and the
moderate amount that
customers are willing to pay has prevented the development of a modeling-upon-
request industry
Many have predicted a burgeoning industry in modeling just around the corner. The
advent of VHDL, for example, led to predictions that numerous companies would soon devote
themselves wholly to the business of developing models. Similar expectations have appeared
from time to time for analog modeling. In neither domain have such expectations been met,
however. In particular, the widely asked-for "model-upon-request" service has
never made much headway.
The reason the modeling industry hasn't taken off is a lack of neither standards nor
technology; it's not even pigheadedness on the part of EDA companies. The real reason is
economics.
Consider the question of modeling a PLL. Creating a reasonably complete model for a PLL
is a large project. It takes a knowledgeable engineer something like four to six months to
create the first model. After that, future models will take roughly two to three weeks
each. Creating a model is a large-enough undertaking that engineers often ask EDA
companies for modeling upon request.
The economics of such a service does not meet customer expectations. Given a (very
approximate) salary of $80,000/year for an engineer experienced enough to create a PLL
model, four months of work to create this model would cost $27,000. Even if it were
possible to sell the same model to several customers, the cost would still be many
thousands of dollars. And that is only the labor cost of creating the model; it includes
none of the many other costs that go into making a final product.
Most customers are willing to pay $50 to $500 for a model. Clearly, a large difference
exists between what customers think is a reasonable price for a model and the cost of
actually developing such a model. The companies that make simulators can and do spread
such costs over many sales of the same simulator. A modeling-upon-request service cannot
spread the costs, because there are so few requests for the same model.
Many people have also suggested creating a "bulletin board" where customers
could interchange models. The Internet and World Wide Web would allow the creation of
sites where people could post models and where others could download them. However, these
sites would do nothing about the underlying economics. It still takes a lot of work to
develop a custom model, and there is still little repetition of part numbers that need
models. In addition, anyone who downloads a model from an open bulletin board has to do
additional work to verify that the model is accurate and complete enough for his
simulations.
The gap between the high cost of developing models and the moderate amount that
customers are willing to pay has prevented the development of a modeling-upon-request
industry. Modeling work primarily occurs in situations in which the company can spread the
modeling cost over many customers (for example, EDA companies that sell simulators). This
economic factor is fundamental and shows no signs of changing. And if the economics
doesn't change, won't change either the likelihood of a modeling industry.
|
