Analog Devices Inc. began microelectromechanical-system development in the late 1980s. Since then, ADI has shipped hundreds of millions of MEMS-based accelerometers for automobile applications, and last year it broke into consumer electronics by shipping more than 1 million three-axis accelerometers for Nintendo's Wii video-game controller. ADI envisions MEMS applications in microphones and acoustics, medical diagnostics and drug delivery, RF devices, and ultrahigh-precision measurement and test equipment. Robert Sulouff, director of business development at ADI's dedicated MEMS facility in Cambridge, Mass., recently spoke with EE Times contributor R. Colin Johnson.
EE Times: Who inspired ADI to jump on the MEMS bandwagon?
Robert Sulouff: It started in the 1987-1988 time frame, when a Boston University professor stopped by and talked with Steve Sherman [the first recipient of ADI's annual Founder's Innovation Award]. From that introduction, we began experimenting with adding MEMS to our integrated circuits.
EE Times: When you invented your accelerometer, were you thinking about the air bag trigger application?
Sulouff: Somewhat, but at that time there was a lot of talk about smart automotive suspensions to control the quality of the ride, and accelerometers were viewed as a good sensor for that. It wasn't until several years later that the mechanical air bags became a dominant problem--just as we were sampling our first accelerometers.
EE Times: So automakers switched over?
Sulouff: That's right. It was one sensor instead of a collection of them around the vehicle. It was an accelerometer instead of a switch. It had the continuous-monitoring reliability. And the system's implementation cost was maybe one-tenth that of the mechanical solution.
EE Times: Can you outline the process steps ADI uses to add MEMS mechanical structures to CMOS chips?
Sulouff: Our strength is fabricating the MEMS mechanical structure alongside its electronics on the same CMOS chip. We first process a bipolar CMOS chip, with typical transistors, using all of the high-temperature processes, and we reserve an area in the center of the die for the MEMS sensor portion. After making this IC, we add the MEMS using a special polysilicon--a thick material that you would normally use to put down CMOS gates.
EE Times: How do you keep from turning the rest of the chip into soup while you are adding the MEMS?
Sulouff: You have to allow for the [high] temperatures. The way we do it in the current implementation is to make large transistors using an older process that can take high-temperature annealing. Our BiCMOS, before it has any metal or anything else on it, can take 1,000°C for three hours.
After we add the MEMS, we go back and finish the integrated circuit by putting down the contacts, the metal and the passivation, and then we remove the sacrificial material underneath the MEMS structure so it's free and clear.
EE Times: So the sacrificial material is removed after the metallization step?
Sulouff: Correct. It's the last step.
EE Times: Is it low-temperature?
Sulouff: Yes, it is. We put down a photoresist, then use hydrofluoric acid for about an hour to remove all the oxide underneath the polysilicon.
The entire process has a collection of patents [associated with it]. Some of the critical ones cover the ability to hold together all those small mechanical structures while you are going through the wet etches. There are two parts of the secret sauce there.
The first is to use a soft photoresist [a photoresist that is softened by ultraviolet light passing through the transparent areas of the mask] to hold all the fingers and structures apart from one another in a wet solution--because if you allow them to just float around, whenever you pull them out of the liquid they stick to one another with surface tension.
The second part is to add an oxygen-plasma dry etch, just to remove the photoresist. And for an extra measure afterward, before we dice up the wafer, we put down a vapor coat, just a few atoms thick, to keep it from sticking while in use.
EE Times: Isn't the wafer capped, to keep out contamination?
Sulouff: Everything we do now is capped. Historically--for over 10 years--we put the die into a hermetically sealed ceramic package, but now we put a cap on it.
We have a second wafer that has a structure with a cavity [for the die] and a ring that goes down on top of the circuit. We use glass that flows and melts at around 400°C. It comes down aligned on top of the regular circuit wafer with the MEMS, where we heat it up so it bonds and forms a hermetic seal.
EE Times: It sounds like you've taken a normal, macroscopic hermetic sealing method and "MEMSized" it.
Sulouff: That's quite correct. We wanted to take it down to the individual die level--like a little clamshell right over the MEMS portion, so you can then treat the die just like any normal die. Because you have protected the MEMS part, you can then wire bond it--put it in a lead frame and shoot it up with plastic.
EE Times: That also has to lower the cost.
Sulouff: That is what we are striving for. It cuts the cost and makes a smaller part, which makes it easier to handle for the new cell phones and videogame controllers.