Inside Intel. Tim Jackson

      Sometimes, understanding why a device failed was as important as finding a way to make it work. One of Intel’s most clever engineers was a tall, thin physicist called Dov Frohman who had devoted most of his working life, including a PhD and a spell inside Fairchild’s research lab, to the study of an obscure metal-nitride oxide semiconductor technology. When Intel decided to bet on silicon gate, Frohman was reassigned to trouble-shoot the process. Looking at the broken devices, he began to wonder whether part of the reason the silicon-gate devices had failed might be that some of the gates in the circuit had become disconnected, or ‘floating’.

      Some weeks later Frohman wheeled a demonstration trolley into Gordon Moore’s office. In the course of ten short minutes he demonstrated a way of turning the phenomenon that caused Intel chips to fail into a something that could form the basis for a new product. ‘We put together a 16 bit array with primitive transistor packages sticking out of the sixteen sockets,’ Frohman recalled. ‘There were red bulbs to indicate the bits. This was all new to us, and we were thrashing around. We showed Gordon that by pushing the button you could program the device, and we demonstrated that it would hold a charge.’ Frohman had discovered something truly astounding: a semiconductor that behaved like read-only memory (ROM), but could be programmed with incredible ease.

      This discovery had the potential to change the lives of engineers who needed to store information permanently on a chip. The standard way to do this was to ‘burn’ the data on to a standard ROM by drawing out the circuit on sheets of rubylith, turning the design into a set of glass masks and then waiting for the circuits to be etched on to silicon wafers on the fab line. Usually the process took weeks, if not months. Now engineers would have a short-cut: they could ‘write’ the data into this new kind of ROM in a few minutes – and the prototyping cycle for a new computer product would be cut from months to hours.

      Moore did not need to see Frohman’s demonstration twice before committing the company to turning it into a commercial product. Soon after the development programme began, however, Frohman came back with some even better news. It wasn’t just writing data to his new kind of ROM that was easy. The chip could be easily erased too, using ultraviolet light. So if the circuit designers found that they had made a mistake, or decided to make some improvements to their design, all they needed to do was to take the ROM chip off the circuit board it was installed on, open up its top, and shine a UV beam on it. The chip would then be like new, ready to receive a fresh set of data. Intel decided to call it an erasable, programmable read-only memory (EPROM).

      Before the EPROM was ready for sale, the company had to find a way of manufacturing it at reasonable cost. The design’s first manifestation, a chip with a capacity of 2 kilobytes (2K) was from this point of view a disaster – it was half as big again as the biggest chip Intel had ever made. Simple arithmetic meant that you’d get fewer devices, or ‘dice’, off each silicon wafer if the chips were bigger, so the average cost of each one would be higher. But there was a further penalty for trying to build bigger devices. One of the key sources of failure came from random flaws in the silicon wafer itself. Other things being equal, a bigger die size would mean that a higher percentage of the devices that came off the fab line would be flawed. The outlandish size of the first EPROM also presented a more prosaic problem. The chip was so big that its design didn’t fit on the layout tables. The circuits had to be assembled from four smaller layouts stuck together – which created alignment problems.