When the programmable read-only memory arrived in the '60s, it was a powerful addition to an engineer's arsenal. But it had a glaring limitation: It could not be reprogrammed. Once a bit map was programmed by blowing thin-film nichrome fuses, the memory could not be changed, zapped fuses could not be reset, the PROM could not be erased. Until Dov Frohman came along.
Working at Intel Corp., Frohman was wrestling with a severe reliability problem as he developed nonvolatile memories efforts he had been pursuing earlier at Fairchild before he joined Intel in 1969. The devices he was developing were unstable. Lines that were not connected anywhere were charging up with no apparent source of charging voltage. And these strange charges were responsible for the instability.
Naturally, he aimed his initial effort at uprooting the source of these charges. Then came the "eureka!" moment. Instead of eradicating the charge, Frohman figured, let's try to use it. Let's use the problem as a solution a solution for creating stable nonvolatile memories.
Frohman designed a MOS device with an extra gate floating above the control gate with no connections. Then, passing a heavy current through the source-to-drain channel, he could drive a charge up to that floating gate a charge that would remain after power was removed. By driving that charge onto floating gates of selected transistors in an array, one created an electrically programmable read-only memory. This was a leap ahead of the blown-fuse programmable read-only memory that had been invented earlier. It was no longer necessary to blast fuses to create a bit map. The EPROM would replace the PROM.
What's more, this EPROM could be erased by exposure to X-rays. This was to be expected because X-rays are a high-energy type of radiation, so an EPROM could be erased within its sealed package. Unfortunately, X-rays posed their own problem. They created oxide instability, so Frohman wasn't happy with X-rays.
But knowing that X-rays worked, however unhappily, Frohman started investigating other types of radiation to excite electrons up to the floating gate and, to erase, by driving them back into the substrate. In time, he came up with the idea of shining ultraviolet light through a quartz window (which gave the EPROM an additional name, UVPROM). UV could indeed erase the device. But it posed another problem: how to convince customers that a package with a quartz window was reliable.
Gordon Moore, company president at the time, came up with the idea of placing some EPROMs on the building's rooftop and exposing them to sunlight for a long time. There was no degradation of the charge on the floating gate. So, while the sunlight test might not have stood up to rigorous scientific evaluation, it was enough to convince customers that EPROMs retained their memory.
In September 1970, EPROMs were demonstrated to the Intel brass. The devices met, first, with skepticism, then wild enthusiasm. They met with similar enthusiasm when Frohman introduced the EPROM in a paper and a film at the 1971 International Solid-State Circuits Conference in Philadelphia. (The presentation was awarded "best paper.")
When Intel started selling the EPROM (the 2-kbit 1702), the expectation was that the chip would be used as a prototyping IC. An engineer would test his system with the bit pattern he programmed in the EPROM, then switch to the less-costly mask-programmed 2k ROM for volume production.
Then, in the spring of 1971, Intel introduced the 4004, the first commercial microprocessor. It set the stage for a vast market for the EPROM. The microprocessor wanted a program, and it would be handy if this program could be altered.
Designed for reuse
The EPROM offered enormous benefits. Since it could be reprogrammed, it could be used repeatedly, even in different systems, so it could speed product development and lower the cost. Further, it dramatically simplified field upgradability.
A few months after the 4004's introduction, in 1972, Intel began selling the 2-kbit 1702, a p-channel EPROM that required three voltages +5, - 12 and - 25 V voltages that weren't available in most systems because they weren't needed for anything except that EPROM.
The part had other limitations. Erasing the 1702 required exposure to ultraviolet light for what might be a long time. Depending on the intensity of the UV, erasure might take half an hour. Further, the 1702 was a p-channel device, so it was much too slow, even in the early 1970s. And its data retention was unreliable.
In 1974, just two years after the EPROM's commercial introduction, George Perlegos joined Intel from American Microsystems and, working for project leader Phil Salisbury, proceeded to make two significant contributions. He developed the first n-channel EPROM, the 8-kbit 2708. This was important because the recently introduced 8080, the first high-volume microprocessor, was an n-channel part, so the n-channel EPROM, faster than its p-channel predecessor, could work nicely with it.
The chip was easier to program, mainly because one could ground the substrate. And it used exactly the same three voltages, - 5, +5, +12 V, that were used in the first NMOS microprocessor, the 8080. A fourth voltage, +25 V, was used only in the external programmer. And at 8 kbits, the 2708 began to approach the density of RAMs which, in 1974, was 16 kbits.
Then, in 1976, when Zilog introduced the 5-volt-only Z80 and Intel introduced the 5-volt-only 8085 microprocessor, Perlegos developed the 2716, the first 5-volt-only EPROM, the first 16-kbit EPROM and the first EPROM to get rid of the three-voltage hassle for operating the device.
Perlegos used ion implantation to adjust thresholds. So -5 V was no longer required for substrate bias and the 2716 required only a +5-V supply. It still required a higher voltage for programming, but that voltage, +15 V, was part of the programmer, not the system being developed. Perlegos was first to employ ion implantation in a programmable device.
In 1978, Perlegos fathered a more dramatic breakthrough, the 2816, an EPROM that could be erased and reprogrammed rapidly without ultraviolet and the quartz window. Further, and just as important, this was the first nonvolatile memory device that could be read, erased and written to just 1 byte or one row of bytes at a time. Thus, unique to this day, it did not require a time-consuming and awkward bulk erase.
By using an extremely thin gate oxide about 100 angstroms instead of 500 to 800 one could tunnel electrons into and out of the floating gate. Applying about 220 V between gate and drain (the substrate) tunneled electrons up to the floating gate. The device, an electrically erasable EPROM, was called an EEPROM or E2PROM.
Then came a move. In 1981, Perlegos; his boss, Phil Salisbury; and Gordon Campbell, marketing director for nonvolatile memories, left Intel to create Seeq Technology. Perlegos was vice president of engineering and Campbell was president. Salisbury became vice president and general manager.
They didn't like the idea of requiring an external 20-V supply to program the E2PROM. Instead, since programming now required minuscule current, Perlegos used a charge pump (an oscillator and capacitor circuit) within the chip to create 20 V for programming. For the first time, it was possible to erase and reprogram the chip remotely without removing it from the circuit and without an external 20-V supply or an external programmer.
And they developed the first 64-kbit E2PROM, the 2864.
Just three years after he was involved in creating Seeq, at the end of 1984, Perlegos left to found Atmel which later acquired most of Seeq.