Figure 1. A household ant lies over a 12-layer nickel microchain with six independently movable links. The height is ~96 µm.
People may one day be using a machine the size of a refrigerator to automatically build microelectromechanical systems (MEMS) with the one-touch ease of operating an office copier.
Researchers at the Information Sciences Institute (Marina del Rey) at the Univ. of Southern California (USC) have developed a first-generation system that fits onto a desktop and is capable of fabricating a 3D microdevice.
The system, which uses a new technology called electrochemical fabrication (EFAB), offers several advantages over existing MEMS manufacturing processes. Scientists using LIGA, surface micromachining, or bulk micromachining processes face several problems in designing their systems. LIGA (an abbreviation of the German words for lithography, electroplating, and molding) is essentially a lithography process that depends on thick layers of photoresist that are exposed using high energies from a costly synchrotron, and it is limited to "2.5-dimensional" single-layer shapes similar to extrusions.
Surface micromachining has expanded the number of layers to five, thanks to researchers at Sandia National Lab. However, the limited number of layers and time-consuming manufacturing process limit device complexity while entailing significant manufacturing costs. The high costs reduce the number of applicable MEMS devices to those niche structures that are important enough to warrant such cost. Finally, bulk micromachining is very limited in the complexity of devices that can be produced.
True 3D design
To bring MEMS into the mainstream, USC researchers believe the emergent industry needs a flexible, low-cost, fast, and easy-to-use manufacturing alternative to the conventional methods above. The technology should be broad enough to fabricate many different types of structures -- from microvessel mixing tanks to miniature motors and mirrors -- and it should be compatible with CMOS semiconductor technology.
With those goals in mind, the USC group, led by Adam Cohen, adapted the principles of solid free-form fabrication (SFF), or rapid prototyping, a standard approach used to create macroscopic models and prototypes in the automotive industry and elsewhere. SFF takes a 3D CAD file and converts it into a prototype model by depositing one layer on top of another, and repeating the process until the full structure is completed.
Like SFF, EFAB offers an unlimited number of layers to the designer. According to Cohen, this allows the designer to work straight from a 3D design file. "When people traditionally do MEMS design," Cohen said, "they're usually working at the 2D level, trying to anticipate a desired 3D shape while designing mask layouts in 2D. They're really working backwards."
In addition, the limited number of layers available with current micromachining processes means the designer has to come up with a functional structure while working within the constraint of just a few layers or cross sections. Cohen's team is currently working on software that will automatically take a 3D CAD file and turn it into a series of 2D patterns on a photomask, enabling one to design in three dimensions.
EFAB requires manual effort to go from the 3D CAD file to the 2D patterns needed for each layer. "Basically, anyone with a PC will be able to design a very complex microdevice on screen by using 3D CAD, importing the geometry into our software, and automatically generating a file that a photomask-making machine requires," Cohen. said.
EFAB uses a special mask called an "instant mask." The mask creation process, however, starts with the same steps used to create a photomask for a microchip. This is the only part of the EFAB process that requires a cleanroom. Just like with LIGA or other lithography processes, the initial photomask creation can take anywhere from a day to a week, depending on the amount of money spent, number of layers, etc. After the initial photomask is created, it is used to fabricate an instant mask formed by a metal anode and a conformable insulator material. This proprietary process takes about two days, Cohen said.
While these first steps resemble the steps used in lithography and other MEMS manufacturing processes, it does offer a few benefits. "Since our substrate is usually much smaller in area than the instant mask by a factor of up to 100, we can put many cross-sections on a single mask," Cohen said. "Then we move the substrate around in the machine and select which mask is required per layer. As the substrate gets larger, you can put fewer cross-sections per mask, but there's no reason you still can't process all the masks you need simultaneously, instead of having to repeat the photolithography over and over again for each layer."
Cohen said the proprietary process also surpasses standard electroplating processes in speed and simplicity. A single electroplated deposit produced using conventional photolithographic methods would require nine or ten steps per layer and between three to five hours. EFAB can produce such a deposit in 20 to 30 minutes and require far fewer steps. Cohen's team is now busy building a compact, self-contained machine to automate the EFAB process.
Figure 2. EFAB process flow. (a) selective deposition of first material using instant masking; (b) after first material deposited; (c) after blanket deposition of second material; (d) planarization; (e) after repeating for several layers; (f) after removal of one material.
'Printing' on the substrate
Once the instant mask is created, the substrate and mask are loaded into the EFAB machine and the EFAB process begins (Figure 2). The substrate is pressed against the mask within an electrochemical bath where a first material (e.g., copper) is deposited. The substrate is withdrawn and is placed into a separate tank where a second material (e.g., nickel) fills in the areas left uncovered by the first material. The layer is polished to a planar surface, and the process is repeated as many times as necessary. Once the requisite number of layers has been deposited, the entire substrate is put into an etching bath where the copper is removed, leaving the free-form 3D structure.
Leaving the copper, which serves as a sacrificial material, on the substrate until the final step means that EFAB can create as many layers as it needs with virtually no geometrical restrictions or damage to delicate structures associated with depositing subsequent layers or polishing. Cohen explained that while multilayer surface micromachining can accommodate up to five layers, the process is extremely time consuming and residual stress in the vacuum-deposited films would make it difficult to achieve structures with minimal distortion if more layers were used. Furthermore, unless planarization is used on every layer, it becomes difficult to use more than a few layers since features become distorted when deposited over a nonplanar surface.
Cohen's group has already made a 12-layer chain, featuring independent links (Figure 1), to demonstrate how EFAB can create multiple-component devices in a single process without assembly. This success does not mean that EFAB has already displaced more conventional methods for MEMS manufacture. Presently, LIGA and surface micromachining offer smaller feature sizes (down to a few microns), while EFAB has so far only produced features of 20 microns in size.
However, Cohen hopes the next system will offer minimum feature sizes of 5 to 10 microns and expects EFAB will eventually match other processes for minimum feature sizes. "But really, we don't see a huge need in the near future to go below about 5 microns for MEMS devices. We're not building semiconductors after all," Cohen said. "Also, the sidewalls of EFAB-produced structures are not as smooth as those made with LIGA due to the multilayer nature of the process, which can be an issue for some applications".
Another drawback is that EFAB, at present, only works with materials that can be electrodeposited (primarily metals, alloys, and some semiconductors), although the USC team is beginning to pursue approaches to incorporating additional materials into devices. Many experts believe MEMS will not really come of age until the structures can be combined on integrated circuits (see "Industry adjusts to the MEMS market," p. 4).
Addressing this issue, Cohen said he believes EFAB can be one part of a two-part process. Although it would require some preparatory steps, EFAB could create MEMS structures on a completed semiconductor wafer. The wafer would have to be protected against some of the chemical baths used in EFAB. However, unlike other MEMS manufacturing processes, EFAB works at comparatively low temperatures, posing no threat of temperature damage of the microchips. (CMOS ICs can tolerate up to 400 degrees celcius. EFAB processing doesn't exceed 60 degrees celcius.)
"We believe that we will be able to deposit arbitrarily complex structures (e.g., inductor coils) on top of a wafer and have them mechanically and electrically interface to the chip," Cohen said. "It's something we've worked out conceptually, but as yet we haven't pursued the IC/EFAB integration. We hope to find the resources to do that soon."
Bump-bonding a MEMS structure to a microchip is another alternative, Cohen said, but it requires additional packaging steps that could potentially increase the manufacturing costs and limit device geometry. "I think people are doing bump-bonding today because their processes are not compatible with CMOS technology," Cohen said. "You can't put a chip in a high-temperature oven."
In addition to the ease of design, complex 3D structures afforded by an unlimited number of layers, and low-temperature processing, another strength of EFAB lies in its speed. Although the time needed to create a photomask and instant mask are comparable to those used in conventional MEMS processes, once manufacturing starts, EFAB quickly pulls away from the competition.
"In less than an hour, we can fabricate a layer that would take surface micromachining one or more days to complete," Cohen said. He said standard lead times for a MEMS prototype at industry prototyping facilities such as MCNC (once called the Microelectronics Center of North Carolina) are eight to ten weeks from CAD file to a prototype of just three layers. EFAB takes up to a week to go from CAD to instant mask, but after that, it can build layers at the rate of 30 to 35 per day.
"Micromachining and other processes might be able to get the first few layers down in about the same time it would take us, so initially we might be neck and neck," Cohen said. "But after that, we're going to pull ahead very, very fast."
University of Southern California
Information Sciences Institute
4676 Admiralty Way, Ste. 1001
Marina del Rey, CA 90292
Tel: (1) 310/822-1511
Fax: (1) 310/823-6714
EFAB Web site: www.isi.edu/efab.
R. Winn Hardin
R. Winn Hardin is a technology journalist based in Jacksonville, FL.