How to Blast Your Chip with High Energy Neutron Beams

So you want to know how reliable your chips are and how susceptible they are to single event effects (SEEs) where a neutron or an alpha particle causes a storage element (flop or memory cell) to flip in a way that alters the behavior of the device. There are two ways a particle hitting a device might not cause a problem. Firstly, the particle might hit an area of the chip that is not vulnerable to particle upsets. Or secondly, an upset occurs but due to other circuitry it doesn’t propagate to an output and thus doesn’t matter.

As with most reliability measures, waiting ten years to find out how reliable a device is over ten years isn’t really very useful. A certain amount can be done in shorter periods and then extrapolating statistically but most of the time this simply takes too long. What is required is a way to accelerate the testing by increasing the rate that the devices are bombarded with particles so that ten years worth of particles can be compressed into a day or two.

There are three types of particles of relevance:

• high energy neutrons (from cosmic rays)
• thermal neutrons
• alpha particles

Alpha particles are stopped by almost anything, just a few centimeters of air for example, so alpha particles in practice have to come from the materials used in manufacture: package plastic, solder, silicon and so on. So alpha particle testing requires the chips to be de-capped or the package would largely absorb them.

Thermal neutrons are generated by a nuclear reactor and so finding it is relatively easy to find a source of low energy neutrons and put your devices in harms way and see what happens.

The most difficult is high energy neutrons. There is no convenient source of high energy neutrons so they need to be made specially at a resource like the Weapons Neutron Research (WNR) facility at Los Alamos. Of course, one immediate question is how do you accelerate a neutron given that at is…well, neutral (A neutron walks into a bar. How much for a beer? For you, no charge.) The answer is that you accelerate protons with a linear accelerator that is a couple of miles long (like the one at the Stanford Linear Accelerator that highway 280 passes over the top of, but much longer). You then put a target in the beam, such as a block of tungsten. This will generate neutrons along with protons and other stuff. So using a powerful electrical field the charged particles can be diverted leaving neutrons. And then using a collimator that is basically a pipe about 12″ in diameter the neutrons that are going in the wrong direction can be absorbed leaving a beam of neutrons.

The beam has a profile of neutron energy that is close to that of cosmic rays, although there are orders of magnitude more neutrons. So if your chip (or system) is put in the beam then it undergoes the effects of years of cosmic rays in a few dozen hours. Since these are single event effects not permanent damage of the device, the chip needs to be powered up and running some sort of test program to pick up the effects when they occur and count them.

As if that doesn’t sound complicated enough, beam time at WNR is booked a year in advance so for an individual company to do this is actually quite risky. If anything goes wrong then you risk missing your beam time slot and having to wait a year for another one.

iROC Technologies does this on a regular monthly schedule so that if you miss your slot with them one month, then you can get a slot the following month. iROC has a lot of experience with this sort of testing, how to set it up, how to execute it, how to analyze the data afterwards. Plus what can go wrong and how to prepare for it so that you don’t waste valuable beam time trying to, for example, find a spare cable.