The evolution and logic behind Ethernet sensor and actuator connectivity
03/01/2001
Paul Sagues
Time and again, the unrelenting progress described in Moore's Law has wreaked havoc with engineering vision. With the benefit of hindsight, we can now see that sensor bus technology in the semiconductor industry has gone in a direction few would have predicted.
 |
Connectivity evolves
About 20 years ago, the automobile industry, in an effort to reduce vehicle wiring, introduced a serial bus designed to multiplex many signals over a single conductor so that left- and right-turn signals, taillights, backup lights, etc., would not need their own individual conductors. The protocol, called controller area network (CAN) bus, was a big step forward from earlier bit-bus technology that used glorified universal synchronous receiver-transmitter (USRT) technology to achieve multidrop capability with serial lines.
About 10 years ago, attempts were made to build relatively low-complexity network chips for very price-sensitive applications, such as office lighting control. Each network chip was given a unique address at manufacture time. Some of these networks, such as Lonworks, attempted to apply the technology to general-purpose industrial control, including semiconductor tools.
Much earlier, Xerox was trying to address the need for high-speed intra- and intermodule communication, and invented Ethernet, a vastly more complex method of sending serial information on a wire.
Against this backdrop, those involved with the industrial controls world tried to set the course of history.
The question they wrestled with was how to connect all digital and analog sensors and actuators used in various industrial processes and machines, including semiconductor process tools. Conventional wisdom dictated that the technology of wiring sensors and actuators to a central programmable logic controller (PLC), first developed in the 1960s, would have to change for the same reason that the automobile industry had invented the CAN bus. The cost and inflexibility of wiring each actuator and sensor was too high.
Thus, each controls giant picked or developed its own serial bus structure. Many jumped on the fragmenting CAN bus, and soon we had CAN, CAN-open, and DeviceNet, in addition to Profibus, Lonworks, Profibus-DP, Seriplex, Interbus-S, and a few oddball niche buses like Sercos. In all, there are about two dozen different serial buses used throughout the industrial controls universe.
Which bus?
Despite enormous investments and periodic declarations of victory by the various bus camps, none has gained a large share of the industrial controls market. Certain semiconductor toolmakers have similarly endorsed various buses, but no bus has gained a significant foothold in semiconductor fabs. Four factors appear to have stalled progress:
· Insufficient bus bandwidth. Most buses run well below 1Mbit/sec. While this bit rate is sufficient for some simple machines and processes, latencies in too many machines render these serial buses ineffective. For example, semiconductor tools contain hundreds of digital inputs and outputs and scores of analog inputs and outputs. Packetizing and transmitting this information at typical DeviceNet rates can easily result in latencies in excess of 0.1 sec, often too slow for closed-loop control or high-speed machine sequencing.
· High price. A single connector for some of the popular buses can cost upwards of $100. Sensor bus interface cards for PCs can easily cost $500-1000. Semiconductor tool builders are under intense pressure to reduce the cost of systems. The added cost for device-level networks often does not offset the promised reduction in wiring cost.
· Lack of interoperability. Since the relatively simple serial bus protocols do not have application connection layers, the application layers are invented by each sensor or actuator vendor. The interoperability sought by the OEM is thus not available; the OEM either buys a sensor with a proprietary application layer or writes its own application layer. A semiconductor tool builder is therefore faced with a substantial software effort to integrate a device such as a valve manifold or a servomotor drive. Given the limited sensor bus bandwidth, software must often be handcrafted and cannot be used from one tool to another.
· The Godzilla factor. Each of these industrial serial buses is backed by an industrial controls titan, and none will concede to the others. The result is a war of attrition in which the OEM machine builder is left without a clear choice. Just because one semiconductor tool builder chooses a particular sensor bus does not mean that its customer will also see it as a plant-wide standard. Tool builders are not unified in their sensor bus selections. The result is a lack of standardization with no prospect for reaching common ground.
Ethernet technology
Conventional wisdom has dictated against using Ethernet as a sensor bus. After all, Ethernet has always been more expensive than simpler serial buses and has involved more complex software. Still, there are compelling reasons why semiconductor tool manufacturers should embrace an Ethernet sensor bus, including:
· Ethernet is about as native to the Windows-NT environment as can be found. Achieving a reliable link into the NT world is essential for modern semiconductor tools to facilitate recipe management, product tracking, and remote e-Diagnostics.
· In general, plant maintenance personnel are ill-equipped to maintain networks. On the other hand, IT and MIS departments at fabs have already embraced Ethernet technology everywhere else in the facility.
While the 10Mbit Ethernet bandwidth is 10x higher than what is available from the other two-dozen industrial control networks, the perception is that it is still too low to handle sensors and actuators in high-performance machines. One can easily argue that Ethernet, given its nondeterministic architecture, needs more bandwidth to provide the same performance as a simpler sensor bus. Semiconductor tools still need millisecond-level I/O performance. For example, the timing of gas valves on deposition tools determines gas volumes. To achieve precision, the valves must respond predictably. However, the perceived limitations of Ethernet have evaporated with the advent of 100Mbit bandwidth. This ten-fold increase in bandwidth puts Ethernet in a unique position to provide the millisecond-level I/O performance that a semiconductor system builder needs.
How has 100Mbit Ethernet addressed the other problems associated with serial buses?
· High price. Thanks to Moore's Law and the adoption of Ethernet by the PC industry, the cost of an Ethernet drop is at least one-tenth the price of much lower performance serial buses. Rather than a $1000 sensor bus interface card for the PC, the 100Mbit Ethernet node often comes free with a PC. Whereas Ethernet was originally much more expensive than device-level buses, today it is much less expensive.
· Interoperability. Ethernet is a mature protocol with well-accepted transport and application layers. Transmission control protocol (TCP), Internet protocol (IP), and file transfer protocol (FTP) are standards in the NT world. Custom software need not be developed.
· Network support. Try asking your MIS or IT group which of the two dozen serial networks they would pick over Ethernet. Indeed, there is strong support by these professionals for extending the Ethernet wire from the office environment all the way to the sensor level. Plant maintenance takes responsibility for the sensors and actuators while MIS and IT assure the integrity of the plant network. Gone are the three or more different networks required to make a tool run.
Support for Ethernet
At the recent Semi International Trade Partners Conference, the use of Ethernet on semiconductor tools was the topic of numerous presentations. One of the most dynamic and convincing was by Howard Charney, VP of Cisco Systems, a staunch proponent of Ethernet as the backbone of the enterprise. It is clear that Ethernet dominates the enterprise. It is also clear that the long-sought link to enterprise resource planning must enable data flow. What better way to enable this flow than by using the same physical wire?
A successful approach has been to deploy a motion and machine control system based upon 100Mbit Ethernet technology in which not only the servo control system communicates over the Ethernet wire, but also the sensors and actuators. The market is demonstrating that OEM tool builders are embracing this open, low-cost, and high-bandwidth medium, given its strong acceptance by the MIS and IT community.
The engineering vision of 20 years ago that was wrestling with simple serial protocols could not imagine a day when Internet protocols moving at billions of bits/sec would exist. But, once again, our technical vision had not factored in Moore's Law.
Paul Sagues is president and co-founder of Berkeley Process Control, 1003 Canal Blvd., Richmond, CA 94804; ph 510/236-3333, fax 510/236-1186.