New Type of Fatigue Discovered in Silicon

Invisible Pink Unicorn writes "Researchers at the National Institute of Standards and Technology (NIST) have discovered a phenomenon long thought not to exist. They have demonstrated a mechanical fatigue process that eventually leads to cracks and breakdown in bulk silicon crystals. Silicon — the backbone of the semiconductor industry — has long been believed to be immune to fatigue from cyclic stresses because of the nature of its crystal structure and chemical bonds. However, NIST examination of the silicon used in microscopic systems that incorporate tiny gears, vibrating reeds and other mechanical features reveals stress-induced cracks that can lead to failure. This has important implications for the design of new silicon-based micro-electromechanical system (MEMS) devices that have been proposed for a wide variety of uses. The article abstract is available from Applied Physics Letters."

Small gears vs. Large gears?

[StickyWidget]

Duh.

Study was conducted on the micro-mechanical objects modeled after mechanical objects in the macro- world. So, in essense, small gears will wear down and break just like big gears do. This isn't really a discovery, all large mechanical devices are subjected to a rigorous set of conditions that they will encounter. Just because a group of scientists never subjected the micro-versions to the macro-equivalent test doesn't mean this is new type of stress, it means that nobody though to check it.

And before anybody posts anything about flash memory or processors, this doesn't apply. Memory and processors are "solid state electronics", not "Micro mechanical devices", and are not vulnerable to the same type of stresses (i.e. those caused by friction, shear, or centrifugal forces).

~Sticky
/Duh

The fatigue scale is all wrong for today's MEMS

[compumike]

They're talking about displacements of hundreds of micrometers... it's not clear that any silicon actually displaces that much under any sort of normal operation. Even in common MEMS parts like accelerometers (like those controlling your car airbag or Wiimote), the displacements are tiny -- typically on the order of one micrometer -- although they do happen hundreds of thousands of times per second.

Ever heard of plastic versus elastic deformation? Elastic is when it's small enough to come back to it's original state (no permanent effect). Plastic is when the material is permanently reorganized. They're at a huge displacement scale, so it's not clear how this applies to modern MEMS systems which are moving two orders of magnitude less.
--
if(coder && wantToLearn(electronics)) click(here); [nerdkits.com]

Re:Small gears vs. Large gears?

[caerwyn]

You didn't RTFA, did you?

The findings are relevant to silicon precisely because the macro-level tests have *not* shown fatigue cracks. Now, the article suggests that this may be a weakness in the macro-level testing methodology, but it doesn't change the fact that silicon was considered "special" because of it's structure, and now it appears not to be.

So, uh, you've actually got this completely backward. No one thinks it's a new type of stress, it merely wasn't expected that silicon would be susceptible to it.

NOT LEDs!!!

[mangu]

LEDs are not made of silicon. They are either gallium arsenide or boron nitride, depending on the color.

Re:DLP TV/Projectors, the first consumer victim?

[StickyWidget]

The answer is no, but it could be subject to other types of mechanical stress. The difference is that the experiment was done to gauge damage from differing direct pressure, DLP use something called a micro mechanical torsion spring. The experiment doesn't quite scale to the spring. However, the way the torsion spring works is that it allows twisting, kind of of like the old "bird in the cage" [bizarrelabs.com] persistence of vision trick. It's designed to accept a degree of stress from the pressure of twisting. Conceivably, if the crystal layers were aligned in a way that put differing stresses on different layers, it could be an issue. Kind of like if you do the bird trick too long you start seeing small bits of thread pop off from the main string.

However, the kind of tolerance is *probably* already present in the DLP chips. The forces that the spring is subjected to were carefully calculated, and the technology has been in use since the 60s. You could probably take a look at the older types of DLPs and compile evidence that a large amount of cycles won't harm it.

Caveat: I am not a micro-mechanical device engineer, but I follow developments. I figure micro-mechanical devices will need control systems of some sort someday.

~Sticky
/It's all about temperature, pressure, and friction.

But it makes an upper bound

[EmbeddedJanitor]

Yes, current MEMS operate at a scale where the effect likely has no impact (probably by a few orders of magnitude). However, previously this effect was thought to not exist and thus was not a factor in future MEMS design. That really only limited MEMS devices to size constraints where there is enough mass etc to provide a measurable effect.

Now that a stress issue has been found that places a limit on how the materials can be used and how much MEMS devices can be shrunk etc.

Re:DLP TV/Projectors, the first consumer victim?

[nullspace]

Actually, Analog Devices probably has a larger MEMS rollout and probably for a longer time. MEMS is incorporated into airbag systems [analog.com] (about 200 million units and the largest market share at around 60%), IBM's Active Protection System for Thinkpads [nanotechwire.com] and of course Nintendo's Wii controller [analog.com]. I would assume that this fatigue would be something worthy of further examination. Disclaimer: I work for Analog Devices, but not as a product designer.

Re:The fatigue scale is all wrong for today's MEMS

[secPM_MS]

There are scale issue here. Even in metals with significant fatigue issues, such as Aluminum, if the structure is thin enough, the image forces on a dislocation suck it to the nearest free surface and you avoid the growth of dislocation tangles that result in fatigue failure. If I remember properly, the relevant thickness for Al was on the order of 100 nm. Note that I am working from memory from grad school ~ 25 years ago, when I did my Ph.D in fracture mechanics.

TI has been working with the mirror systems for a long time now, I suspect on the order of 20 years. They should have real reliability info to work from.

Airbag sensors are the highest volume MEMS...

[StandardCell]

...and as for DLP, it's a valid question especially given that they oscillate rapidly thousands of times a second to simulate brightness levels (they're pulse width modulated to full reflect or full absorb mirror positions). However, the NIST abstract says that their test is done with a spherical indenter presumably imparting impulsive loads of some magnitude. I don't know how big the sphere is or what material it's made of since I don't have the full article, but I'll assume it's some microscale silicon ball; hopefully they didn't do something like ceramic shattering glass easily with little force. DLP stresses would normally be torsional stress along the micromirror hinge of a magnitude dependent on the deceleration at the limit of the DLP motion and the mass of the mirror. Now, if TI was clever and didn't modulate the mirror past the elastic limit of the material, they might be able to largely overcome this problem. Cantilever-style micromirrors might not fare as well because the material is always being deformed, though I again assume they do a stress-strain plot to ensure they don't go past the elasticity limit. On that note and to come full circle, one would assume that sensors do not exceed their ductile elasticity limit except in critical situations, such as high shock as is found in an abrupt movement of an accident. Then again, they're typically single-use.

Re:oh well... a more expensive alternative

[pimpimpim]

When you write with a pencil you move graphite layers around, from the pencil to the paper. They stay graphite layers, however. The point is in the layers, the layer-layer bonding is weak pi-bonding, which makes it easy to detach the layers from each other by shearing them (e.g. writing).

The point why diamond stays like it is, is that even though it's thermodynamically unstable, it is kinetically stable. In contrast to graphite, it is very hard in diamond to break all bonds between all atoms in the lattice. The chance of this happening is also so small because you would need to break various bonds at the same time, and statistics goes down. In the case that this would happen, however, diamond would be most likely turn into the lower-energy graphite shape.

Another point of view from which diamonds are hardly "forever" is the resale value. A diamond, just like most champagne, isn't really worth much, the worth of a diamond drops tenfold the moment you've put your signature on the receipt. The only reason it is expensive is because they are marketed like that. There are some excellent articles about the marketing around diamonds on the internets.

Re:NOT LEDs!!!

[JollyRogerX]

Wrong. The first LED was made of Silicon Carbide back in the early 1900's. An LED made out of straight silicon would produce infrared light. The reason they don't make infrared LEDs out of silicon today is that they are not as efficient (i.e. they produce too much heat). You can show this is true with Schroedinger wave equations.

Not "New Type"...

[florescent_beige]

It's old fashioned fatigue, and it isn't new. This [cwru.edu] paper quotes (2nd para) 1992 work that demonstrated fatigue in micron-sized silicon specimens.

Silicon is a typical low ductility material that does not tolerate cracks very well because there is very little plastic deformation at the crack tip (the process zone). Fracture mechanics is based on an energy balance, when the amount of energy absorbed by the creation of the fracture surfaces (the surface energy) plus the amount of energy required to do that plastic work in the pz is equal to the amount of strain energy in the structure that's released when the crack gets bigger (the strain energy release rate), the crack becomes unstable and the part goes bang.

The strain energy release rate varies with the load and crack size, for a given crack size at loads lower than the critical load, pre-existing cracks (there are always cracks even if they are microscopic) open a bit and the pz deforms. When the load is released, the pz doesn't go back to it's original configuration. Repeating the apply-load remove-load cycle progressively grows the pz which causes the crack to get bigger in some complicated ways. But think of it this way, the crack tip is theoretically infinitely sharp (the limit is the inter-atomic distance of the material). This discontinuity causes infinite theoretical stress which causes the atomic bonds to break at the tip. Process zones have been the subject of countless PhD theses.

In a low ductility material the energy absorbed by the pz is small compared to the energy absorbed by the surfaces created when the crack grows. Remember the pz is responsible for fatigue growth, the pz plus the surface energy is responsible for unstable crack propagation. So a small pz means you have to load the material close to the crack instability load to get fatigue growth. With a small enough pz it's impossible to load the material accurately enough to grow the crack without breaking the part. So THATS what they mean by silicon being immune to fatigue.

It seems like the reason this is not the case in microscopic silicon specimens is another PhD topic, the explanation is complicated. Oxidation caused by humidity in the air is a factor, as well as loading in the compression mode.

Again, all this has been known for many years.