How The Computer Got Into Your Pocket

Today’s high-tech innovators know that simply creating a clever bit of technology is no guarantee that it will be adopted. In many cases an invention does not become accepted until it is used in a “killer application”—something irresistible enough to overcome people’s natural reluctance to change. Thirty-five years ago Texas Instruments (TI), with headquarters in Dallas, faced the same problem. In 1958 TFs Jack Kilby had become one of the inventors of the integrated circuit (IC), more familiar in its present form as the chip. Today chips can be found in everything from coffeemakers to computers, but in the early 1960s few engineers entertained any thoughts of using them. Almost all integrated circuits went into military applications, and while these could be quite k profitable, the general public was a much vaster market. In 1965 Pat Haggerty, president of TI, wanted to start placing his company’s ICs in consumer products. To do that, he needed to invent something enticing that could only be made with ICs.

In the early 1950s TI had used a similar strategy to create a market for transistors, which are essentially small electronic switches. Although the company had developed a transistor that could be massproduced, most manufacturers stuck with vacuum tubes. Haggerty and his colleagues responded by developing the technology that made possible the Regency transistor radio of 1954. That pocket-size radio used just four transistors, but it started an avalanche of transistorized products. Haggerty knew that ICs would take off the same way if he could create the right application.

Haggerty shared his idea with Kilby, who ran the company’s semiconductor research and development laboratory. Kilby recalls: “One day I was going to the East Coast on an airplane, and by chance Pat Haggerty was on the same flight, so we sat together. He wanted to do something to increase the visibility and sales of integrated circuits, and he thought that it needed a dramatic product for that. He suggested three. A hand-held calculator was one. He thought that a recorder the size of a lipstick that would record for thirty minutes would be a good deal. And frankly, I’ve forgotten what the third one was. Of the three, the calculator was the only one that looked like it was reasonably realizable.”

Calculating machines were nothing new, of course, but in the mid-1960s the smallest one took up about as much space as a typewriter. The Sharp Corporation introduced the first commercially available all-transistor desktop calculator in 1964. Even though it could only add, subtract, multiply, and divide, it weighed 55 pounds, had to be plugged in, and cost $2,500. Most engineers stuck with slide rules instead.

Haggerty imagined a calculator that would run on batteries and be small enough to hold in the user’s hand. Such a device could never have been built with vacuum tubes, because it would have been too big and required too much power. A vacuum-tube calculator could possibly have been carried around in a wheelbarrow, along with a car battery. Even the latest transistors could not make a hand-held calculator. The Regency radio could fit in a person’s hand because it had used only four transistors, but a calculator would need hundreds. Only ICs could allow the necessary miniaturization.

ICs solve problems with binary logic, which means that the signals inside the circuits come in one of two states: a high voltage or a low voltage, often called a 1 or a 0, respectively. Combinations of transistors create devices called logic gates. A simple logic gate examines the states—0 or 1—of two inputs and generates a predetermined output, also in the form of a 0 or 1.

A variety of logic gates exist. In an AND gate, for example, the output is 1 only if both inputs are 1. A NAND gate works just the opposite: Its output is 1 unless both inputs are 1. In essence, logic gates are switches—just like transistors, except a bit more complicated. Combinations of logic gates can solve many kinds of problems, including arithmetical ones.

Haggerty knew that including all the logic gates needed for a four-function calculator would result in a very complex IC, much more so than anything that had been made at that time. And designing the circuitry would be only the first step, for even the simplest ICs were notoriously difficult to manufacture. “In the early 1960s,” Kilby says, “a respectable yield for a single transistor was on the order of 10 or 20 percent, so the thought of doing a dozen of them on a chip seemed overwhelming. Even by ’65 the thought of doing hundreds of them on a chip seemed overwhelming.” This was a serious problem, since in contrast to the forgiving cost structure of military contracts, the calculator was expected to sell for less than $100. But Haggerty’s persuasive enthusiasm made Kilby and his team feel that they could overcome any obstacle.

As with most corporate projects, Kilby kept his calculator effort under wraps. That meant giving it a code name. A previous project at TI had required engineers to do some work in Boston, and they had given that project the code name MIT, even though it had nothing to do with the Massachusetts Institute of Technology. When another secret project came along, it seemed natural to call it Cal-Tech, even though it had no connection with that institution or anything else in the state of California. As a code name, Kilby later said, “it was a miserable choice. Anybody who heard it would have figured out that we had a crash project going on calculator technology.”

To get Cal-Tech started, Kilby needed a project manager. In September 1965 he picked Jerry Merryman. Merryman had been at TI for only two years, but in that time he had developed a reputation for solving complex problems. Merryman recalls, “The first I knew about this project, Jack called me and two other engineers into his office. He said that he and Pat had been talking about the pervasiveness of electronics and how someday it might be possible to have a personal computing device that you could put in your pocket and hold in your hand, and of course it would have to work on batteries. He had a little book on his desk, and he said that it would be nice if it was no bigger than this book. He said it would have to have some buttons or something for you to tell it the problem and some neon lights or something to tell you the answer.

“I don’t believe we ever used the word calculator in the project at all,” Merryman continues. “Kilby always described it as the slide-rule computer.” (This nomenclature provides a curious turnaround from the early days of digital computers, when such machines were invariably called calculators.) “When we were having this conversation, if you went out on the market to buy integrated circuits, the most complex one had about twenty transistors in it. This thing that was going to be a calculator was going to have about eight thousand transistors, but we didn’t know that yet. It was a fair leap.”

That leap would require more than just a mountain of improvements in IC design and manufacture: The Cal-Tech team had to invent nearly every aspect of the device from scratch. They needed a way to get information in, a way to make calculations on that information, a way to get the information out, and a way to power all these processes. At the start they didn’t have a solution to any of this. Nevertheless, Kilby expected his team to solve everything, and quickly.

At the first Cal-Tech meeting, Kilby requested an overall layout of the project in just three days. Merryman says, “I went back to my blackboard, and I thought, Well, what is it that you have to do? I thought, I guess it’s got a button on it somewhere. Maybe it’s got a plus sign on it, and if you push that button, it was going to add some stuff. And I just chicken-tracked that all over my blackboard for three days and nights.”

From the beginning he decided to encode information in a form called excess-3 binary code. In this system a base-10 digit (0 to 9) is converted to its binary equivalent plus 3. For instance, the number 1 is 0001 in regular binary and 0100, or 4, in excess-3 binary. This form of coding simplifies some mathematical operations by creating a turnover, or “carry,” when the excess-3 codes total 16, meaning that the underlying decimal digits total 10. This allowed the Cal-Tech to operate on decimal digits one at a time. Without excess-3, the binary total and the decimal total would have turned over at different figures, making calculation much more complicated. Merryman also decided to rely entirely on NAND gates because he recalled hearing that you could make anything with them. They were fairly simple to design; he could make a NAND gate with just eight transistors and two resistors.

During his 72-hour designing burst, Merryman laid out essentially all the logic—the electronic network—that would drive the Cal-Tech. He figured out, in at least a general way, how a user would enter a problem and how the Cal-Tech would add, subtract, multiply, and divide. He also developed a system of electronic components that would direct the flow of information throughout the sliderule computer. Despite its hasty preparation, his design endured as the project progressed. According to Kilby, “Jerry was probably one of the few people who could have done that at that time.”

While the project got rolling, Kilby and Merryman kept it secret, even within the walls of Texas Instruments. “It used the vast resources of a large company, as it had to, but there were only very few people that we directly involved in the project who knew that we were building a calculator,” Merryman says. In fact, “my wife didn’t know that I worked on the calculator till years after they were on the market.”

Kilby did add one more main player, James Van Tassel, who had been working for him. “My job was basically to make things,” Van Tassel says. “I had fabrication capabilities.” The Cal-Tech team needed fabrication. For one thing, they needed a way for the user to communicate with the electronics. In today’s keyboard-infested society, it’s hard to imagine a time when compact electronic keyboards did not exist. Yet at the time, Merryman says, “there were no good examples of cheap, compact keyboards. We had to have a keyboard that was thin, simple, cheap, reliable- work a million times.” Kilby turned that challenge over to Van Tassel.

He started with a piece of Mylar, a thin, flexible plastic film produced by Du Pont. He clad the film with a layer of copper and then etched most of it away, leaving only quarter-inch dots beneath the spots where the keys would be. Depressing a key would push a dot of copper plating against a circuit board below. Instead of clicking like modern keyboards, the prototype felt somewhat mushy, but it did the job. By landing on a set of lines on the circuit board, the dot of copper connected some of the four outputs of a register to ground, making them 0s. The untouched lines remained 1s. Those four binary digits represented an excess-3 number.

Consider the 5 key, for example. The decimal number 5 is 0101 in binary and 1000 in excess-3 binary. Pressing the 5 key would connect the three least significant bits in a register to ground, making them O and leaving the most significant bit a 1. The digit thus created was stored in an electronic way station, where it waited to enter Merryman’s logic circuits.

While Van Tassel worked on the keyboard, Merryman got under way making Cal-Tech’s ICs. The finished calculator would contain four of them. With a system of such complexity Merryman might have been expected to test his logic design before etching it in silicon. But he didn’t. He says, “I wrote the logic out on a piece of paper and checked it and checked it and asked two other engineers to check it.” When they gave their approval, the Cal-Tech team went right ahead to chip production.

Merryman recalls, “I probably felt as scared as a guy walking across a tightrope on Niagara Falls who had previously done a ten-foot one in his barn.” He laughs and adds, “I was not a highly experienced person at that kind of designing, but I was an adaptable person, and I did it. If I were to go back and look at it now, or if some knowledgeable person were to go back and look at it now, you’d say, ‘Well, he did some of that pretty stupid. You could cut out 3 percent of the parts over here, if you wanted to take a different tack or something.’ But that was really kind of a side point.”

Much of the trouble with putting lots of transistors on an integrated circuit came from the poor reliability of the manufacturing process. The more transistors a circuit needed, the more likely it was that some of them would not work, and with fabrication techniques still at an early stage, the chances that a complicated chip would come out error-free were very slim. Yet when Merryman took his circuit layout to TI’s chip-making facilities, he told them he wanted an 83 percent yield—that is, he wanted 83 percent of the eight-transistor NAND gates in a chip to work. Merryman recollects, “They all fell on the floor laughing, saying, ‘We’ve never seen anything better than 25 percent.’ ” When they asked Merryman how he expected to get such a high yield, he said, “I’m going to be using low-power circuits and very wide tolerances in the voltages and currents that’ll work and simple layouts and wide conductors and so on.” Merryman also increased the odds of getting a usable chip by putting on 384 NAND gates when he needed only about 150 of them.

Still, the chips came out unreliable. “None of those circuits were going to work,” Merryman says. “They were just far more complex than anything ever made before, so each one really had a little something wrong with it. You had to find out what it was and then fix it.” Finding the flaws turned into a project in itself. “If you made an integrated circuit that was an inch in diameter and had thousands of transistors on it and 122 connections coming out of it, and you had never seen anything like that before, and that piece was a fraction of something—it didn’t do the whole job—how would you test that thing and say that piece is okay?”

To test the new ICs, Merryman built an enlarged version of Cal-Tech’s logic. Using about 200 commercially available ICs and lots of wire, he and his colleagues built a calculator circuit that sprawled across the top of a lab bench, requiring a two-tiered stand to hold it. This circuit operated exactly like the Cal-Tech and could be used to test a chip. Isolating one chip at a time, Merryman would run a simple test signal through it and the desktop circuit and compare the results. If they came up with different answers, a light would flash. It flashed with every chip they tried.

To locate the exact problem in a faulty IC, Merryman and his colleagues would put it under a microscope and trace out the connections with mechanically positioned probes. Once they found a problem, they either removed a little bit of the metal or added a tiny piece of wire to jump over some defects. After lots of microscopic surgery they got about half the chips to work adequately. These integrated circuits, like modern ones, solved problems in steps. The pace of the steps was determined by a clock that coordinated them. The Cal-Tech ran at about 50 kilohertz, or 50,000 steps per second—some 6,000 times slower than the clock rate in a typical modern computer.

The calculating circuits inside the Cal-Tech relied largely on adding. To add two binary numbers with electronic circuits, a series of logic gates starts by adding the least significant bits: 0 plus 0 makes 0, of course; 1 plus 0 or 0 plus 1 makes 1; and 1 plus 1 makes 0 and carries 1 to the next addition. In the Cal-Tech, Merryman designed such an adder with about 60 NAND gates.

The Cal-Tech used the adder to accomplish all other arithmetical operations. To subtract, the Cal-Tech first inverted the number being subtracted, so that 0s became 1s and 1s became 0s. In excess-3 coding, this yields the 9-complement, or 9 minus the number that was inverted. Adding together the 9-complement, the number being subtracted, and 1 results in the required difference. For example, to calculate 8 minus 5, first convert both numbers to excess-3 binary, so that the problem is expressed as 1011 minus 1000. Next invert the second number, yielding 0111. Then add 1011 plus 0111 plus 0100 (the excess-3 code for 1) and discard the part that goes over four digits. This yields 0110, which is the excess-3 code for 3.

Numbers greater than 9 were operated on one digit at a time, borrowing and carrying when necessary, just like in pencil-and-paper arithmetic. Numbers with a decimal point were treated as whole numbers, with the decimal point restored at the proper place after the calculation was complete. To multiply, the Cal-Tech simply added repeatedly. To divide, it subtracted repeatedly until the result was 0 or negative, keeping track of how many subtractions were needed. This might seem cumbersome, but input numbers could be a maximum of only six digits long, with an output up to twelve digits. Since the calculation circuits were very fast, even the division of a very large number by a very small number required less than a second, with the bottleneck taking place in printing the output, not in processing the numbers.

Merryman’s design divided the Cal-Tech’s logic into four main chips, which were made from round pieces of silicon a little less than an inch in diameter. Each one contained lots of tiny transistors that were connected to one another with a rectilinear web of tiny metal lines. These chips had to be placed in some sort of package that would protect their fragile structure and provide communication with the rest of the Cal-Tech. The machine’s complicated logic would require an unheard-of 122 such connections, or leads. “Most packages at that time were fourteen or sixteen leads, so this had to be a significant departure from past packaging approaches,” Van Tassel says.

He made the first package from a square piece of circuit board. The silicon wafer sat inside a circular hole. The hard part was finding a way to make connections between the 122 leads on the silicon wafer and the 122 lines that radiated across the package to the other three chips or to the rest of the Cal-Tech—keyboard, printer, battery, and so forth. To make these connections between chip and package, Van Tassel used so-called flying leads: gold-plated copper leads that were cantilevered from the package and ultrasonically bonded to the wafer. Ultrasound was not a standard way to attach things at the time, but in a case of availability being the mother of invention, TI happened to have an ultrasonic bonding device on hand, and as Van Tassel says, “We just scrounged that up.” The technique did not always make good connections right away, and Van Tassel had to experiment to find the proper temperature, amount of ultrasonic energy, and amount of gold to put on the leads.

Once the developers had found how to input and process a calculation, they still needed a way to display the result. A few years later the first generation of commercial calculators would use light-emitting diodes (LEDs), but in the mid1960s such devices were just becoming practical. Moreover, developing a continuous live readout would have added another hurdle to the project—the last thing they needed. It would also have boosted the price, and since it would leave no record of a calculation, it could have been less useful than a permanent display. So the Cal-Tech team decided to stick with the tried and true technique of a paper printout.

An ink-and-ribbon system would have been too cumbersome for a hand-held device. Fortunately, some TI engineers were working on a thermal printer, and Kilby suggested that Merryman take a look at it. The idea was promising, but putting it into practice required numerous fixes, large and small. “The thermal printer is quite a problem,” Merryman says, “because the way it prints is kind of like a branding iron. This little spot of silicon has to go up to about two hundred seventy degrees Celsius, and it does that in a time of about ten milliseconds.” A three-by-five array of silicon heater elements created patterns of dots, printing numerals on wax-coated paper.

One particularly difficult problem cropped up in the seemingly simple matter of advancing the paper. The wax on the paper tended to stick to the silicon heater during the “branding” process, and as Merryman says, “You had to jerk that sucker off there. … I think somebody had told us that you could pull it off with 65 grams of force. That would have been more nearly right if they’d said 150.” After making a detailed study of “the energy relations and the different kinds of steel and the annealing and the air gaps and the mechanical relations and the inertia and so forth,” Merryman and his coworkers came up with a reliable device that printed about 12 characters a second on a quarter-inch-wide strip of paper.

Another part of the Cal-Tech that was simple in concept but difficult in execution was the power source. Even today, Merryman says, “If there’s anything that still needs to be invented in the world, it’s a battery.” Three decades ago the possibilities looked even bleaker: “Nickel-cadmium batteries were just being mentioned, but they weren’t any good yet.” The developers settled on silverzinc instead. But then during the project new, more reliable kinds of nickel-cadmium batteries came on the market. Merryman and his colleagues switched to these before completing the Cal-Tech. To supply its thousands of transistors and power-hungry thermal printer, they packed the Cal-Tech with eleven 1.5-volt rechargeable batteries, which provided about four hours of use.

In thinking back on all the obstacles the Cal-Tech team faced, Merryman said, “It would seem reasonable to most people, I’d think, to solve those problems in series. We’ll solve problem A, and then we can more carefully design problem B and then problem C.” But that’s not how they approached it. Instead, Merryman and his colleagues took on all the problems in parallel. While Merryman was working with the chip makers on the ICs, Van Tassel was designing the keyboard and packages for the chips. During all that, Merryman also found time to work on the printer and batteries. In some cases, they had more things going on than they could do in a day’s work. In developing the recharging system for the batteries, for instance, Merryman said, “I built the charger for it in my own garage.” In the end, all the different pieces came together.

Finally they had a usable device. Merryman says, “Probably late in 1966, maybe December 1966,1 probably was the first guy in the world to hold a box in my hand with batteries in it, with no wires attached, that calculated.” That box came from a solid block of aluminum that had been milled out to hold all the batteries, the printer, and the ICs. The case was 4.25 inches à wide, 6.15 inches long, and 1.75 inches thick. The I entire Cal-Tech weighed 45 ounces.

Kilby and his colleagues applied for a patent on September 29, 1967, two years after the beginning of the project. On June 25, 1974, after twice revising and refiling the application, they were granted patent number 3,819,921. Kilby estimated that the entire effort cost TI a few hundred thousand dollars. No exact figures were compiled, because the project never had a budget. “The expenses were spread out over a period and could be folded into other R and D,” says Kilby. It was money well spent.

TI was not yet ready to embark on a full-scale manufacturing effort, so it shopped the Cal-Tech around the electronics industry, looking for someone to make a commercial product from its prototype. Canon licensed the technology and created the Pocketronic, essentially a replica of the Cal-Tech. It went on sale in Japan in the fall of 1970 for the equivalent of $395 and in the United States in February 1971 for $345. According to Guy Ball, a calculator historian and collector, “The Pocketronic weighed about a pound and a half, but it was a portable unit. You could—in theory—stick it in your big pocket. It would weigh you down a little bit.”

Soon after the release of the Pocketronic, the hand-held calculator market exploded, and the Wild West period of calculator manufacturing was under way. In the fall of 1971, using TI-built components, Bowmar brought out the smallest four-function calculator yet, the first widely available model that could comfortably fit in most pockets. It had an LED display and sold for just $240, making it temporarily the cheapest model as well. The following summer TI introduced its first self-manufactured handheld calculator, the four-function TI-2500, or DataMath, priced at $150. Dozens of other firms began selling calculators around the same time, and the competition was fierce. By the end of 1972 a four-function model could be had for $100; by 1976 the price was under $20; and by 1980 the four-function calculator had become a virtual giveaway item, retailing for $10 or less.

At the other end of the market, in January 1972 HewlettPackard introduced its HP-35, a “scientific” calculator that went well beyond simple arithmetic to allow for logarithms, exponents, trigonometric functions, and memory. Hewlett-Packard, which had been making electronic desktop calculators since the mid-1960s, developed its own technology for the purpose and did not license any TI patents. Ball says the HP-35 “had a waiting list of close to six months, and it was a $400 unit.” But its features quickly became a standard package, the first level above the basic four functions. In 1976, when Texas Instruments introduced the best-selling calculator of the 1970s, its scientific TI-30, the price was just $25. Meanwhile TI, Hewlett-Packard, and other companies developed increasingly sophisticated calculators for scientists and engineers, with programming capabilities and scores of tiny keys. Such calculators virtually eliminated the use of slide rules, dominating the field until they were themselves superseded by personal computers. Advances in technology have allowed today’s hand-held calculators and computers to do much more with smaller battery requirements than the Cal-Tech. More efficient logic designs requiring fewer circuits have saved power. Also, today’s calculators use liquid-crystal displays, which create no heat or light and thus consume virtually no power.

A collection of Cal-Tech parts, essentially a nonworking case and keyboard, is preserved at the American Computer Museum in Bozeman, Montana. The Smithsonian Institution’s National Museum of American History owns the original Cal-Tech, the only working model, which Kilby presented to the museum in 1975.

Looking back on the project, Merryman says, “You had all these things that were difficult. You had no keys. You had no batteries. You had no low-power transistors. You had only high-power printing means. You didn’t know how to make a calculator.” Reflecting on how his team maintained its confidence, he adds: “It’s a little bit being keyed up. It’s a little bit being scared, but also it’s sort of having faith—based on your previous experience—that things are going to work out.” And work out they did, not just for the Cal-Tech project or Texas Instruments but for the entire concept of integrated circuits—and thus for the electronics industry as a whole.