I gave a talk with the same title as this blog at the TSensors Summit held in La Jolla, California on November 12-13. The ‘T’ in TSensors stands for Trillion Sensors and the TSensors Summit initiative is addressing the provocative question: what will it take to get to a worldwide market of a trillion sensors a year in the not-too-distant future, say 10 to 15 years from now. The TSensors initiative is being spearheaded by serial MEMS entrepreneur Janusz Bryzek who cites the book Abundance by Peter Diamandes and Steven Kotler as inspiration for TSensors. The key premise behind the book is that technology is advancing at such a fast rate, exponentially in fact, that we have the opportunity to provide abundant food, clean water, renewable energy and health care for everyone on earth within a generation. This is heady stuff, especially compared to the doom and gloom that pervades the daily news (if only political and cultural differences were as easy to resolve). Sensors of all types will play a key role in technological solutions to these pressing worldwide challenges.
At TSensors events, including this one, speakers have provided many examples of ways that sensors are key to abundance. For instance: low-cost biometric sensors and labs-on-chip could dramatically lower the cost of healthcare; miniature spectrometers could track food, water and air quality; moisture sensors in every crop row could be used to optimize and dramatically reduce agricultural use of fresh water; and so on. It’s clear that a lot more sensors will be needed to attain abundance, therefore one of the main aims of the TSensors initiative is to accelerate development of new sensors.
This brings me to the subject of my TSensors talk: how modeling can accelerate development of new sensors. In MEMS as well as any other fabrication technology one can imagine, sensor development involves a cycle of designing, building and testing. For CMOS and MEMS, this can be described as a “silicon learning cycle”. The problem is that silicon learning cycles are both time consuming and costly. The proposition is simple: if actual building and testing can be replaced with predictive modeling that takes much less time, sensor development time can be dramatically accelerated. I identified three kinds of modeling that together can provide a virtual learning environment for sensor development: device modeling, system modeling and process modeling. I then provided an example of each type of modeling from Coventor’s experience.
I’ll mention here just the device modeling example which demonstrates our own exponential trend in technology. Since 2001, we’ve benchmarked our MEMS simulation software (Architect3D in 2001 and MEMS+ since 2009) on the same transient simulation of a MEMS gyro. Over the years, we made many improvements to our algorithms and implementation, and also benefited from faster computers enabled by Moore’s Law. The graph below shows how the simulation time decreased dramatically over the years as a result of our efforts, from three and a half hours in 2001 to less than 7 seconds today. That’s an exponential rate of improvement of more than 2X per year! That rapid improvement rate is more than twice what can be explained by Moore’s Law alone, which predicts doubling of transistor density (and presumably compute power) every two years. Here’s the thing: I simply did not appreciate how much improvement we’d made until I looked at the historical trend. This type of exponential improvement in technology is exactly the underlying driver for the optimistic forecasts of Abundance. It wasn’t until I took a long-term historical perspective that I realized that Coventor has its own great example of the exponential technological advancement.