The economics of consumer electronics is driving ever smaller semiconductor feature sizes, also known as “nodes.” The node size determines a transistor’s minimum gate width. The latest node is 14 nm – only 140 times the size of a hydrogen atom! At this size, only a hundred electrons or so distinguish between a 1 and a 0. For embedded systems, the march to ever smaller nodes should have ended long ago. Unfortunately, embedded system engineers have little or no control over what nodes are in production.
This is illustrated by comparing Apple Computer to John Deere. Last year, Apple had revenue of $227 billion and John Deer had revenue of “only” $35 billion. Considering all the wonderful tractors and machinery that John Deere produces and how important they are to our food supply, is there something wrong here? Which company should be driving embedded system and Internet of Things feature sizes?
Of primary concern is the steady stream of cosmic rays from outer space. As they collide with atoms in our atmosphere, they generate a continuous background of energetic neutrons. Because neutrons are uncharged, they easily pass through barriers such as buildings, vehicles, and enclosures, but occasionally one collides with a silicon atom in a transistor. When this happens, a shower of charged particles are released, potentially causing a “bit flip,” converting a 1 to a 0 or vice versa. The result is that data, a pointer, or an instruction may be altered substantially, thus leading to an equipment malfunction or failure, such as “unintended acceleration.”
The origin of cosmic rays is not well understood. They are believed to come from supernovas and similar cataclysmic events in the past. Undoubtedly, there are regions in our galaxy where cosmic rays are more dense and possibly more energetic. If the earth were to pass into such a region, the resulting cosmic ray storm could last for days, weeks, years, maybe even decades. It could produce a much larger flux of neutrons that might penetrate the atmosphere more easily. Our cyberspace data-storage cloud could be moved underground or under thick slabs of concrete. What about the billions of exposed IoT devices installed around the world during the next decade? How could they be protected?
Even if a cosmic ray storm never materializes, we have a problem. Deploying the IoT will significantly increase the neutron target. Large numbers of devices will be installed at higher altitudes and latitudes, where neutron fluxes are greater. The steady march to lower nodes will increase semiconductor vulnerabilities. And to make matters worse, parts of our infrastructure, which are currently safe, will be moved into the target area. Catastrophes are sure to follow. We should not be complacent about this threat; our smartphones, tablets, and laptops already fall victim to high-energy neutrons more than we realize. The time to be concerned about this problem is now, before large quantities of devices are deployed.
Where am I going with this discussion? Well, there are things we can do to reduce the future vulnerability of the IoT to high-energy neutrons:
- Stay with current semiconductor nodes or even go back to larger ones for IoT devices. The IoT will produce enough semiconductor sales to break the current economic dependency on consumer electronics.
- Resist the siren song of abundant hardware resources at “low cost.” It leads us to bloated software, which greatly increases the neutron target. Use “right-size” software.
- Start measuring your devices’ neutron resistance. In the long-run, this is as important as cost and power use and should be part of the same equation.
- Make software more fault-aware and self-healing. This also helps to increase malware and bug resistance.
- Move non-essential features into smartphones, tablets, and the cloud, where failures are more tolerable.
See? There’s nothing to be afraid of.
Ralph Moore, President and Founder of Micro Digital, graduated with a degree in Physics from Caltech. He spent his early career in computer research, then moved into mainframe design and consulting.