MEMS makes the trip into a CMOS fab


baolab-metal-mesh.jpgMicromachined chips promise much, which is why every time they turn up in a new system — such as the Nunchuk controller in the Nintendo Wii or the motion sensor in the iPhone — it’s tempting to herald a new dawn for MEMS. It’s invariably a case of “this time it’s all going to happen for MEMS”.

But some big problems still face MEMS. It’s not as cheap to make as you’d expect and the one thing you’d expect manufacturers would have down to a fine art — integration with other microelectronics — is still not easy to do. It’s even hard to package the things. They often need to be carefully sealed using special caps to stop moisture disrupting their delicate inner workings.

The seven-strong team at Barcelona-based startup Baolab reckon they have an answer to at least some of these problems. Having already developed a novel type of MEMS structure — basically a floating bar in an electrostatic box — several years ago, the company was faced with the problem of making it commercially viable.

Dave Doyle, CEO at Baolab, explained that CTO Josep Montoya founded the company in 2003. “He was working on this idea of contactless switching. He established some contacts with some of the big guys who came in and said that the cost had to come down a long way.

“There has been talk in the MEMS world of whether it is possible to build MEMS in CMOS. But what’s new is that someone has figured out how to do the work. It came out of necessity,” said Doyle.

“It is 100 per cent compatible with CMOS. That is really the point. A lot of people talk about MEMS and CMOS in the same breath. Normally they mean MEMS on top of CMOS.”

In this case, the MEMS elements are made out of the same metal wires as those used to connect transistors on a CMOS chip. “As far as I know, Baolab is the only company to figure it out,” said Doyle.

Dave Doyle, Baolab CEOAs some of the patents covering the process have only been filed recently and have not been published, Doyle didn’t talk in detail about the process. But he said both the MEMS design and the processing are important to making this work on a conventional CMOS process. “It is 80 per cent design and 20 per cent process,” Doyle claimed.

“This is bringing together three disciplines which is what has held people back. You have to have a deep understanding of the chemistry that goes on during etch. You have got to understand the CMOS process and how to design devices for MEMS knowing that CMOS was never meant to be used in this way.”

And Doyle stressed that the manufacturing up to the point where etching with hydrofluoric acid releases the moving elements is conventional CMOS processing. The box that contains the MEMS elements is defined using features and shapes that help control the extent of the acid-vapour etch, which can be difficult to control.

“Hydrofluoric acid needs water as a catalyst but once it starts it turns into a self-catalying process that runs away in a second if you are not careful. We developed a way to control that or, more importantly, to stop it,” said Doyle. “Without that control, you can end up with unpredictable cavity size from one wafer to another. We can’t have that.”

To form the cavity, the chip goes all the way through the fab to the point where it has the final passivation layer deposited on top. The one extra bit of the Baolab process then takes over, opening up holes over the box that contains the MEMS element and etching away the intermetal dielectric. Once that step is complete, the holes are filled and the device sent for packaging.

The release step can be performed at the wafer fab or the packaging house. However, Doyle said it’s likely that the wafer fab will take care of this because packaging houses do not like working with hydrofluoric acid. Who does? (Although a lecturer from university did pour neat HF on his hand as his party piece. As with chips, it doesn’t start eating away until enough water is present to kick-start the reaction. Eventually, enough water is drawn from the skin to start the process. But you’ve got a minute or so to leave it there before you have to worry about it.)

Baolab claims, once the holes in the top have been filled in, the chip can use regular epoxy packaging — not the specialised caps that a lot of MEMS devices needs, which helps drive the cost down further. “We use standard deposition equipment to close off the holes,” said Doyle.

The original MEMS device patented by Montoya was a relay switch. This remains one of the key products for Baolab, although there is a strong chance the first one will be a motion sensor. Whichever product comes first, it will be aimed at mobile phones. The company is planning on having engineering samples ready by the end of the year having made test devices on a 0.18µm process running on an 8in wafer line.

The RF switch needs close contact with electrodes, which demands clean contacts which are achievable but the motion sensor does not have that requirement, reducing the technological risk. “Plus, sensors is an established market,” said Doyle. “If we can take a slice of that market because the device is cheaper, you don’t need a lot of market share to generate a lot of cash.

“Because the technology is applicable to a wide range of markets and applications, we have to pick our fights. The greatest volume is in mobile handsets. But there are other consumer products where the technology could be applied,” said Doyle.

Although Baolab will sell its own products first, licensing to other companies is an option for the future. The current generation of technology works with aluminium metallisation as that is prevalent on 0.18µm-class processes. However, because the release step focuses on the dielectric rather than the metal, copper is an option for the future using post-0.13µm technologies.

“Our whole drive has been about bringing down the cost as far as can be physically achieved. The jury is out on copper as to which — 0.13µm, 90nm or another — will be the most cost-efficient process. But we are working toward copper. There are some interesting attractions to using copper, especially on the RF side,” Doyle explained.