The victory of the losers, or the story of the creation of FLASH memory (part3)
We were very happy and proud; you know, young engineers with their own hands were able to make the first “flip” chip. After some discussion between Gordon and Andy, Gordon said: “Guys, you have done a good job, but we will move on towards creating a monolithic chip.” Since I did not have other urgent projects, they also entrusted me with work on this chip.
The development of a monolithic microcircuit had its own problems. After creating a test template for the topology of the future microcircuit, it was necessary to transfer it to the production line in production. The problem was that at that time we had only two production lines: silicon MOS transistors and bipolar transistors. It was practically impossible to create transistors using MNOS technology on our equipment at that time. Nine out of ten microcircuits had defects. After a month of work, I was just desperate, but continued to do the work – to create this chip.
Reference. The figure shows the design of the MNOS transistor (metal – silicon nitride (Si3N4) – silicon oxide (SiO2) – semiconductor). If there is a charge in silicon nitride, opening the transistor is not possible. The use of an MNOS transistor as a memory element is based on this principle.
MNOS Transistor Design 1 – metal shutter; 2,3 – source and drain areas, respectively; 4 – silicon substrate.
About three months after the start of my work, there were problems with our main products, namely with 256-bit memory INTEL 1101 NMOS RAM. We could not sell a batch of these products, because the operating parameters of the microcircuits were unstable. When testing microcircuits at a temperature of 85 ° C and a humidity of 85%, their operating parameters “floated” and lost stability. At about the same time, the manager called me: “Okay, stop fooling around with this MNOS, the real problem is finding out why the INTEL 1101 chips are losing reliability, and the most important thing for us is to find a solution to fix this.” Of course, I immediately started to perform this task.
It turned out that the problem was the dielectric used, it became conductive in conditions of high humidity. A number of errors were made in the production technology, which led to the fact that metal atoms entered the dielectric layer in the microcircuit.
Modern topographic images of the surface of a single crystal of silicon oxide obtained using atomic force microscopy. On the left is silicon oxide that has no defects; on the right – with defective areas (metal impurities).
Under normal conditions for testing the microcircuit, all parameters remained normal, but with increasing humidity, the dielectric obtained the properties of a semiconductor, and affected the stability of the microcircuit. Having understood what the matter was, the decision came by itself – a strict control of the conditions of deposition of the dielectric was needed. A few weeks later, my proposal was implemented in production.
It is difficult to remember exactly when, but it was after the completion of work to investigate problems with defects related to the memory of INTEL 1101, I got the idea to use this phenomenon. Its essence was as follows. When I investigated the dielectric, I noticed that the metal impurities that were in the dielectric could retain charge. Knowing well the MNOS technology, I began to develop and research the first double-gate transistor. Later, the second shutter was called “floating shutter”, since it was virtual. Using MNOS technology, by deposition of silicon nitride (Si3N4), I created a region in the transistor that had no connection with other regions (isolated region) – this was a floating gate. Moreover, the oxide layer between all regions of the obtained transistor had the same thickness and was created using the same technology. The main goal that I pursued during the development is to get a device that will have high reliability. Suddenly we realized that we could create an information storage device. Although silicon nitride (Si3N4) has better dielectric properties, compared to silicon oxide (SiO2), we chose the latter. The reason for this transition was precisely the fact that at that time MNOS technology, which used silicon nitride, had problems in production (the same problems that I encountered when working with a monolithic memory chip).