Industrial IoT and the complexity of semiconductor fabrication

Over the past couple weeks I’ve been looking at the importance of standards to industrial IoT as well as examining the complexity of implementing IoT in industrial settings. Last week I discussed IoT concerns at power plants. This week we’ll have a look at one of the most complex manufacturing feats out there—the microsecond motion timing and coordination required to produce semiconductors.

In the US, industrial machines are an over $200 billion market and the efficiency of the machines themselves are critical in that they contribute to ultimate productivity in a globally competitive market.

Importantly, industrial machines differ from consumer products in that they are customized to the task at hand. Given that automation must often be engineered for the end user, a great deal of precision in terms of timing and architecture is required. Today’s machines are software driven, networked and often include sensors, cameras and the ability to move in multiple axes of motion.

Similar to power generation, many end users in the industrial production sector have cut operational costs by reducing maintenance staffs and shifting reliance on maintenance services to the machine builder. This typically involves remote reporting and advanced warnings so that a technician only services a machine when necessary.

In the case of semiconductor production, manufacturing involves very precise processes in order to create layers of transistors with specific operating characteristics. Chemical and photolithographic steps are used to harden an exact representation onto a silicon wafer. Wafers are then cut into individual chips and electrical contact points are added.

As semiconductor fabrication processes reach 22 and 14 nm, manufacturers are able to pack more chips on a single wafer. Cuting and dicing of the wafer requires precision measuring on the scale of a thousandth of a millimeter. Blades and lasers are used to accomplish this etching and they function in multiple axes of motion and must integrate feedback about positioning at high resolution.

Additionally, delivery of control data between sensors and controllers has to have a latency of less than 100 microseconds. Add to this complexity the reality that cameras and video are increasingly being integrated into manufacturing and they have high bandwidth requirements. Bandwidth limitations at any step in the system create problems and if maintenance and diagnosis is to be handled remotely, connectivity into the lowest layer of a machine must be secure and real-time.

Costs for this sensitive process run very high, and it’s one of the reasons that there are few semiconductor manufacturers. Very few companies have the capital, which often runs into the billions, to build out a facility capable of chip fabrication. It’s arguably also why a company like ARM chose to become a chip designer and focus their expertise exclusively on design rather than an Intel, which is able to both design and fabricate its semis.

And yet, semiconductor manufacturers are reaching increasing levels of sophistication in networked and machine controlling (often remotely) of a chip fabrication process. In the future, as these processes push for higher bandwidth and lower latency on a converged network that will be asked to absorb new traffic as manufacturing processes evolve, there’ll be increasing pressure on these systems.

I’ve noted it before, but we’ll need highly time sensitive networking capability along with standards to aid in machine to machine communication. Bandwidth reservation and standardized communications protocols are also requirements. These are the types of IoT advancements we’ll continually need to driver operational efficiency at the manufacturing level.

But I also see that in the next few years we’re likely to see standards materialize that allow more than just highly capitalized semiconductor manufacturers to implement IoT and advanced machine control techniques in their processes. Why? The payoffs in terms operational efficiency are just too big. And as analytics improve, and perhaps even come down in cost, the case for a highly networked and automated industrial push will grow.

We should look to the semiconductor industry as an example of what’s possible. The difference, of course, is that other industries often don’t have the degree of capital available to the semiconductor industry. But as standards improve and industrial IoT matures, the returns for less sophisticated industrial processes should also creep up.