Who Invented the Intel 1103 DRAM Chip

The World's First Available DRAM Chip

The newly formed Intel company publicly released the 1103, the first DRAM – dynamic random access memory – chip in 1970. It was the bestselling semiconductor memory chip in the world by 1972, defeating magnetic core type memory. The first commercially available computer using the 1103 was the HP 9800 series.

Core Memory 

Jay Forrester invented core memory in 1949, and it became the dominant form of computer memory in the 1950s.

It remained in use until the late 1970s. According to public lecture given by Philip Machanick at the University of the Witwatersrand:

"A magnetic material can have its magnetization altered by an electric field. If the field isn't strong enough, the magnetism is unchanged. This principle makes it possible to change a single piece of magnetic material – a small doughnut called a core – wired into a grid, by passing half the current needed to change it through two wires that only intersect at that core."

The One-Transistor DRAM

Dr. Robert H. Dennard, a Fellow at the IBM Thomas J. Watson Research Center, created the one-transistor DRAM in 1966. Dennard and his team were working on early field-effect transistors and integrated circuits. Memory chips drew his attention when he saw another team's research with thin-film magnetic memory. Dennard claims he went home and got the basic ideas for the creation of DRAM within a few hours.

 He worked on his ideas for a simpler memory cell that used only a single transistor and a small capacitor. IBM and Dennard were granted a patent for DRAM in 1968.

Random Access Memory 

RAM stands for random access memory – memory that can be accessed or written to randomly so any byte or piece of memory can be used without accessing the other bytes or pieces of memory.

There were two basic types of RAM at the time: dynamic RAM (DRAM) and static RAM (SRAM). DRAM must be refreshed thousands of times per second. SRAM is faster because it does not have to be refreshed.  

Both types of RAM are volatile – they lose their contents when power is turned off. Fairchild Corporation invented the first 256-k SRAM chip in 1970. Recently, several new types of RAM chips have been designed.

John Reed and the Intel 1103 Team 

John Reed, now head of The Reed Company, was once part of the Intel 1103 team. Reed offered the following memories on the development of the Intel 1103:

“The "invention?" In those days, Intel – or few others, for that matter – were focusing on getting patents or achieving 'inventions.' They were desperate to get new products to market and to begin reaping the profits. So let me tell you how the i1103 was born and raised.

In approximately 1969, William Regitz of Honeywell canvassed the semiconductor companies of the U.S. looking for someone to share in the development of a dynamic memory circuit based on a novel three-transistor cell which he – or one of his co-workers – had invented. This cell was a '1X, 2Y' type laid out with a 'butted' contact for connecting the pass transistor drain to the gate of the cell's current switch. 

Regitz talked to many companies, but Intel got really excited about the possibilities here and decided to go ahead with a development program. Moreover, whereas Regitz had originally been proposing a 512-bit chip, Intel decided that 1,024 bits would be feasible. And so the program began. Joel Karp of Intel was the circuit designer and he worked closely with Regitz throughout the program. It culminated in actual working units, and a paper was given on this device, the i1102, at the 1970 ISSCC conference in Philadelphia. 

Intel learned several lessons from the i1102, namely:

1.    DRAM cells needed substrate bias. This spawned the 18-pin DIP package.

2.    The 'butting' contact was a tough technological problem to solve and yields were low.

3.    The 'IVG' multi-level cell strobe signal made necessary by the '1X, 2Y' cell circuitry caused the devices to have very small operating margins.

Although they continued to develop the i1102, there was a need to look at other cell techniques. Ted Hoff had earlier proposed all possible ways of wiring up three transistors in a DRAM cell, and somebody took a closer look at the '2X, 2Y' cell at this time. I think it may have been Karp and/or Leslie Vadasz – I hadn't come to Intel yet. The idea of using a 'buried contact' was applied, probably by process guru Tom Rowe, and this cell became more and more attractive. It could potentially do away with both the butting contact issue and the aforementioned multi-level signal requirement and yield a smaller cell to boot! 

So Vadasz and Karp sketched out a schematic of an i1102 alternative on the sly, because this wasn't exactly a popular decision with Honeywell. They assigned the job of designing the chip to Bob Abbott sometime before I came on the scene in June 1970. He initiated the design and had it laid out. I took over the project after initial '200X' masks had been shot from the original mylar layouts. It was my job to evolve the product from there, which was no small task in itself.

It's hard to make a long story short, but the first silicon chips of the i1103 were practically non-functional until it was discovered that the overlap between the 'PRECH' clock and the 'CENABLE' clock – the famous 'Tov' parameter – was very critical due to our lack of understanding of internal cell dynamics. This discovery was made by test engineer George Staudacher. Nevertheless, understanding this weakness, I characterized the devices on hand and we drew up a data sheet. 

Because of the low yields we were seeing due to the 'Tov' problem, Vadasz and I recommended to Intel management that the product wasn't ready for market. But Bob Graham, then Intel Marketing V.P., thought otherwise. He pushed for an early introduction – over our dead bodies, so to speak. 

The Intel i1103 came to market in October of 1970. Demand was strong after the product introduction, and it was my job to evolve the design for better yield. I did this in stages, making improvements at every new mask generation until the 'E' revision of the masks, at which point the i1103 was yielding well and performing well. This early work of mine established a couple of things:

1.    Based on my analysis of four runs of devices, the refresh time was set at two milliseconds. Binary multiples of that initial characterization are still the standard to this day.

2.    I was probably the first designer to use Si-gate transistors as bootstrap capacitors. My evolving mask sets had several of these to improve performance and margins.

And that's about all I can say about the Intel 1103's 'invention.' I will say that 'getting inventions' was just not a value among us circuit designers of those days. I am personally named on 14 memory-related patents, but in those days, I'm sure I invented many more techniques in the course of getting a circuit developed and out to market without stopping to make any disclosures. The fact that Intel itself wasn't concerned about patents until 'too late' is evidenced in my own case by the four or five patents I was awarded, applied for and assigned to two years after I left the company at the end of 1971! Look at one of them, and you'll see me listed as an Intel employee!"