1966

Dynamic RAM

Robert H. Dennard (b. 1932)

From the beginning, small memories limited what computers could do. Called random access memory (RAM) because any location could be read or written at any time, memory was fantastically expensive. And because programmers frequently did not have enough memory to work with, they had to split their programs and data into segments, copying one segment into memory, processing it, and then saving the results on tape.

Enter semiconductors. Smaller and cheaper than core memory, semiconductors were the obvious next technology to use for RAM. IBM assigned electrical engineer Robert Dennard the task of designing its next-generation electronic memory system. Dennard’s original approach used six transistors to create an electronic switch, called a flip-flop, to store each bit. But halfway into the project in 1966, Dennard realized that he could make the devices even cheaper by storing each bit in a capacitor—a device that stores charge—and using a single transistor to alternatively store the charge and read it back when the data was needed. However, there was a catch: capacitors leak their charge away. The solution was to refresh the bit, perhaps a thousand times every second, by reading the bit and writing it back. Because the charge would be constantly in motion, Dennard described his invention as dynamic random access memory (DRAM).

IBM was committed to finishing the design that used six transistors—a design that today is called static random access memory (SRAM)—so Dennard pursued the DRAM as a side project. Finally, in 1967, IBM filed for a patent for DRAM, which was granted in 1968.

The company that ended up commercializing DRAM wasn’t IBM, but Intel, using a less-efficient three-transistor design licensed from Honeywell, an American technology company. Released in 1970, the Intel 1103 was the first commercially available DRAM. It stored 1024 bits and offered both a better price and better performance than magnetic core memories.

Since then, the storage of DRAM has increased as the size of transistors has shrunk. By the early 1990s, manufacturers were putting a million bits of storage on a chip; by the 2000s, a billion bits. Today the rate of progress has slowed, and modern DRAM chips can hold “only” 4 billion to 32 billion bits.

Dynamic RAM memory chips assembled on a dual inline memory module (DIMM), a format widely used in modern laptops.