1970 · Dynamic Semiconductor Memory · Work 859 of 1,871
A work table in the computer room of an electronics works in California, in the foreground a memory card of green laminate with two rows of small oblong white ceramic packages two and a half centimetres long, each with a raised plain smooth unmarked white lid and a row of short copper-brown legs along each side, beside the card a removed square frame of magnetic core memory, a dense weave of very fine copper wires threaded through thousands of tiny dark rings, a hand in a pale blue shirt sleeve laying a single white package onto the woven wires, cold fluorescent light from above, grey steel cabinets soft behind.
What it was used for
The device came out of a contract development for Honeywell. According to the Computer History Museum's notes of an interview with Joel Karp, Honeywell's development manager Bill Regitz came to Intel late in 1969 with a memory cell of 3 transistors; the part for Honeywell was called 1102, it worked poorly, and in the end Intel persuaded Honeywell to switch to the 1103 developed alongside it. Intel's introductory advertisement of 12 October 1970 promises memories of 50,000 to 10,000,000 bits at lower cost than cores. Documented installations: the Burroughs B1700 and B3700, and the MS11 memory of DEC's PDP-11/45, whose G401 module is populated with 1103-type devices, 4096 words of 18 bits, 350 ns access time. Refreshing required 32 read cycles every 2 ms.
What was new
With 3 transistors per cell instead of 6 in static memories, a chip carried as much as 4 times the data, in Karp's account - which is what made semiconductor memory an alternative to core memory, whose price was then falling from 5 to 1 cent per bit. It was not the first dynamic memory, and the one-transistor cell was older and not used here: Robert H. Dennard's patent 3,387,286 was filed on 14 July 1967 and granted on 4 June 1968, and its first claim claims a cell consisting of one field-effect transistor and a capacitive storage element. According to the Computer History Museum, Four Phase Systems had dynamic memories of 1024 and 2048 bits in 1969. In December 1970 Advanced Memory Systems advertised that it was first to deliver a 1024-bit RAM - an advertising claim aimed directly at the 1103. The cell was presented at the ISSCC in February 1970; the product announcement appeared on 12 October 1970.
What the sources leave open: The often-quoted phrase about the crisis of the 1103 appears neither in the Computer History Museum, nor in the Karp notes, nor in Intel's 1971 annual report, nor in Intel's company chronicle; it is not documented. Quantity and value of the Honeywell order are missing, as is the part's share of revenue. The 1103's die area is not documented. The Karp notes are recollections from 2007.