Ian M. Ross
AT&T Bell Laboratories
Shortly after arriving from England to join Bill Shockley's organization at Bell Labs in March 1952, I was asked to arrange a laboratory session for an April symposium on transistor technology. In 1947, Shockley's team had invented the transistor in an effort to replace relays and vacuum tubes in the telephone network with faster, more reliable solid-state devices. My lab session was meant to allow the attendees, most of whom had never seen a transistor, to measure the characteristics of the device touted as having no failure mechanisms and nothing to wear out.
AT&T had long prided itself on its stringent requirements for network reliability. For example, switching machines were to have no more than 2 hours of downtime in 40 years—a policy statement that has had a powerful impact over the years. The requirement may seem somewhat quaint today: Why not 1 hour of downtime in 20 years or 3 minutes per year? I believe the intent was to specify two things—first, the minuscule percentage of time it was acceptable for a machine to be down, and second, that the switch should be designed to serve for 40 years. Two days before the start of the symposium, New Jersey experienced its typical one-day transition from winter directly to the heat and humidity of summer. In those days Bell Labs was not air-conditioned, and when I got to the lab that morning I found that my transistors for the session had lost all their electrical characteristics—the CRT traces were flat! I had discovered for myself what many people already knew: The transistor was sensitive to its environment and particularly to humidity.
The lack of reliability of early transistors was a huge setback and embarrassment to the semiconductor community. The transistor had been lauded as a device with no failure mechanisms, with nothing to wear out. Instead, we had a severe reliability problem—and it took another 14 years to solve the problem completely. Given AT&T's reliability specification, the delay effectively paced the large-scale introduction of semiconductor devices into the telephone network.
Two decades later I was put in charge of the Network Planning Division at Bell Labs. About this time the technology was at hand to permit the conversion of the networks to all-digital operation. Here again AT&T's reliability specification strongly governed the design. The new electronic switching machines were to be controlled by a stored-program processor—in effect a special-purpose computer. The only way we at Bell Labs knew how to meet the downtime requirement was to use dual processors, simultaneously running identical programs. Fortunately, the machines that were introduced in the early 1970s are still providing service today. So far so good!