|
Size: 6892
Comment:
|
Size: 12690
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 52: | Line 52: |
| .'''p109''' [[ moores-law-electronics.pdf | April 19, 1965 Electronics Article ]] .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' .'''p''' |
.'''p109''' [[ attachment:moores-law-electronics.pdf | April 19, 1965 Electronics Article ]] .'''p111''' Rice and chessboard story, "halfway down the board, he has already committed to all the rice in China". Perhaps loose with "halfway", if 50K grains per kilogram, 200kg per million grains, then 2^33 grains (squares 1 to 32) is 170 metric tonnes. Current production is 200 million tonnes, 20 more squares; historically, production might be 1/16th of that, so 16 more squares and 2 more rows. I'd call that 3/4 of the way down the board. The exponential analogy is apt, but the description is sloppy. The author (a reporter) should test his statements (reporters rarely do). .'''p112''' ''"memory chips capable of storing more than one trillion bytes"'' ... ''nope.'' In 2014, a dense memory process might be 20nm linewidth, and a 4 level/2 bit memory cell might be 40 by 40 nanometers. One trillion bytes is 4 trillion 2 bit memory cells, 2 million cells on a side (not counting periphery and driver cells). 2 million times 40 nanometers would be a chip more than '''8 centimeters''' on a side ... probably a lot more. Again, the reporter/author should learn to use a calculator. No citations; [[ https://en.wiktionary.org/wiki/POOMA | POOMA? ]] . [[ https://en.wikipedia.org/wiki/Moore's_law | From wikipedia:]] ''In 2019, Samsung produced a 1 TB flash chip with eight stacked 96-layer V-NAND dies, along with quad-level cell (QLC) technology (4-bit per transistor), equivalent to 2 trillion transistors, the highest transistor count of any IC chip'' that's a terabyte, but 8 stacked chips. [[ https://www.electronicsweekly.com/news/business/samsung-makes-1tb-flash-module-2019-01/ | This article ]] mentions 16 layers, not 96. As of 2020/03, Samsung has just introduced a 512GB product. === Intel === .'''p133''' 1969 Intel first product 3101 64 bit Schottky TTL SRAM, 2x faster than Fairchild's . "didn't leak power" ?huh? 525 mW [[ attachment:INTLS0327-1.pdf | datasheet ]], 16 pins, 4 bits, 4 address, 35 to 50ns . [[ http://www.righto.com/2017/07/inside-intels-first-product-3101-ram.html | recent teardown ]], also 74S289 .'''p136''' 1969 1101, 256 bit static PMOS silicon gate SRAM [[ attachment:1101.pdf | datasheet ]] 8 address din/dout 16 pin 1.5us . October 1970 1103 DRAM, 18 pin 10 address 16V 300ns access time 580 ns cycle time 9um PMOS .'''p148''' Busicom, [[ https://en.wikipedia.org/wiki/Marcian_Hoff | Marcian "Ted" Hoff 1937 ]] .'''p150''' 12 chips 3K-5K transistors each .'''p153''' Noyce told Hoff go ahead, Grove not told,risking Intel .'''p158''' Hoff on road pleasing customers, Masatoshi Shima architect, Frederico Faggin helps finish 4004 .'''p159''' Faggin bootstrap gate .'''p160''' 4001 2K ROM (Stan Mazor) 4002 320 bit RAM 4003 10 bit I/O register 4004 CPU .'''p162''' Feb 1971 [[ https://en.wikipedia.org/wiki/Intel_4004 | Intel 4004 ]] 16 pin 740MHz 4 bit 10um 2250 transistors .'''p166''' 1103 destructive read bug resembled core memory problem .'''p169''' Microsystems International Canada (MIL of Bell Canada) second source .'''p177''' 1201 for Computer Terminal Corporation (CTC, now Datapoint) becomes 8008 - 800KHz .'''p184''' 8008 announced April 1972, Intellec MIC-8 design tools sales take off .'''p192''' Bill Davidow marketing, Dov Frohman invents EPROM .'''p200''' 1979 Intel 8086 bested by Motorola 68000 .'''p205''' Peter Drucker "Nobody pays for a 'product'. What is paid for is satisfaction" .'''p207''' Project Crush - sell customers complete environment, not just best CPU .'''p211''' IBM young Turk Don Estridge +5 engineers to skunkworks at Boca Raton .'''p219''' 8051 (1980) cloned by 20 chip companies, "single most popular product in history" .'''p221''' Meritocracy, executive berated wife to bring cheap car, not the Mercedes .'''p225''' In 1990s, estimated every dollar of founder's stock returned $270,000 (1968 to 1995 59% growth per year) . 1971/10/13 IPO price 23.50 per share, 350,000 shares, 2 cents after 13 subsequent stock splits, $50.08 on 2020/09/04, mkt cap 213B .'''p230''' 1969, Noyce and 28yo mask designer Barbara Maness employee 43, divorced Betty 1974 .'''p239''' 1972-1975 Gordon Moore's "$15M" Microma watch .'''p247''' Faggin started Zilog 1974, nonperson at Intel until 2009 and documentary ''The Real Revolutionaries'' (not available) .'''p251''' In 1975 Wozniak used a $20 6502, couldn't afford an 8080 because Jobs withheld $3K from an Atari consulting contract .'''p268''' Grove memo to Moore 'Intel should contract with you not to go on vacation - the world invariably seem to turn into a piece of shit while you're away." .'''p270''' iAPX 432 fail ''then 960MX, and BiiN failures -> i960, not mentioned'' .'''p279''' 1974 November Noyce marries Ann Bowers .'''p282''' HP's Ed Hayes quality graph, Japanese >90%, US 60% to 70% .'''p283''' para 3 last sentence "Moore's '''haw'''" ... ha ha .'''p290-p348''' Andy Grove biography, mention's his book ''Swimming Across'' .'''p344''' Wrote "Physics and Technology of Semiconductor Devices" in 1967 while at Fairchild .'''p345''' 1996 "Only The Paranoid Survive", strategic inflection point: radical shift or die. .'''p351''' gray market, discarded chips resold, organized crime and theft of scrap .'''p355''' 1974 incomplete Aloha 4 fab guarded by Doberman "Aloha Doghouse" .'''p356''' October 1981 to March 1982; 125 Percent Solution - two extra hours without pay, 125% solution sweatbands .'''p358''' March 1982: 80186 and 80188 microcontrollers, 82586 coprocessor, and 80286 .'''p370''' Craig Barrett Copy Exactly 1985, Fab 5 (Aloha) technological development .'''p372''' yield to 80%, equipment utilization to 60% .'''p378''' Noyce to SIA .'''p390''' 1985 Grove and Moore direct Intel to exit memory business, cut 7200 jobs (1/3) and closed 7 factories .'''p393''' 1985 80386 [[ https://en.wikipedia.org/wiki/John_Crawford_(engineer) | John H. Crawford ]] .'''p396''' Intel vs AMD, then NEC .'''p399''' Noyce mentors Steve Jobs .'''p402''' 1987 Sematech, $100M for 5 years, TJ Rodgers (Cypress) disparages .'''p406''' 1988 Intel improves Toshiba's flash memory, two decades until Samsung outcompetes .'''p412''' 1992 June 2 death of Bob Noyce .'''p419''' 1999 April i486/80486 $950 50MHz, flash $500M business by mid-1990s |
The Intel Trinity
How Robert Noyce, Gordon Moore, and Andy Grove Built the World's Most Important Company
Michael S. Malone 2014 BvtnLib 338.7621 MAL
Some interesting factoids. Some may be incorrect. I may add a few notes later, but I would prefer to do so from a revised edition with less redundancy, more explanation, and citing more sources (many of this book's references are Malone's other books, not primary sources, NOT useful).
I enjoyed Leslie Berlin's excellent The Man Behind the Microchip, and Grove's autobiography and management books.
I want to know more about Gordon Moore ("Trinity" was aggravatingly sparse) so ordered a used copy of Moore's Law: the Life of Gordon Moore by Thackray, 2015 through Alibris.
p4 David Packard managing HP during WW2 (Hewlett served in Signal Corps). Established targets, trusted (female) employees to learn how to meet them.
p5 Lockheed Sunnyvale largest employer in late 1940s
p12 Shockley Semiconductor: Photolithography Bob Noyce and Jay Last, Diffusion Gordon More and Jean Hoerni, Crystals Sheldon Roberts, 2n696 double-diffused-base transistor Victor Grinich and Murray Siegel, investors Eugene Kleiner.
p14 visionary Noyce presentation to Sherman Fairchild
p18 1959 trade show TI integrated circuit, led to Fairchild ICs with planar transistors
p21 error: Hired manager Ed Baldwin left in 1959 (with Fairchild IP, and the mesa process, but NOT Hoerni's planar process) to start Rheem, which failed in 1961.
p22 Charles Spork, Don Valentine (later VC, Something Ventured), Jerry Sanders, Mike Markkula, Jack Gifford (Intersil, AMD, Maxim), Bob Widlar.
p24 Marshal Cox
p25 Andy Grove UCB PhD Chemistry, hired 1964
- Andy Grove's 1967 text "Physics and Technology of Semiconductor Devices" earned me my first summer job at Tektronix in 1969(?).
p26 Lester Hogan at Motorola, later to Intel in 1969.
p27 (cite?) Spring 1965 conference, Noyce announced all major Fairchild ICs for one dollar
p31 error: Widlar sheep, not goat
p32 1966 Fairchild, 12000 employees, $130M sales, Spork and Widlar left for National Semiconductor
- Andy Grove's 1967 text "Physics and Technology of Semiconductor Devices"
p36 Fairchild belittled Noyce, Noyce hired Hogan to replace himself (huge salary and options)
p44 1968 Noyce, Moore, Arthur Rock start Intel, p46 Grove (as employee) follows Moore
p48 Grove recruits Les Vadász Malone obsesses about Grove's dislike of Noyce
p54 Silicon gate MOS
p55 500K × $5 shares convertable debentures, interest-paying IOUs convertable to stock at IPO
p57 $100K investment in 1968 would be $3 billion in 2014
p61 Electronics News January 11 1971 Don Hoefler dateline "Silicon Valley"
Robert Noyce
p62 Noyce born 1927/12/12 Denmark (SE) Iowa father Ralph Congregational minister, Oberlin wife/mother Harriet
- Oldest brother Donald Sterling Noyce May 1923, 1948 Chemistry prof at UCB, died November 2004
p65 Ralph to Atlantic Iowa, then Decorah, then 1938 Webster City (demomination's associate superintendent), then Grinnell
p70 12yo Bob and older brother Gaylord (1925-2009) build a glider
- youngest brother Ralph Harold Noyce
p74 Odd jobs for Grinnell College physics professor Grant Gale (1903-1998) (who also taught Herbie Hancock)
p84 graduates Grinnell 1949, MIT 19xx Professors John Slater and Wayne Nottingham
p88 musicals lead performer, meets Betty Bottomley9 1930-1996) at Tufts
p90 A Photoelectric Investigation of Surface States on Insulators
p92 Philco R&D under Bill Bradley
p93 patent for germanium thickness measurement during etching. rTansmission intensity drops (interference?), then rises until desired thickness is reached. But Philco was going downhill.
p97 1955 APS paper "base widening punch through" read by William Schockley. Noyces moved to California Feb 23, bought house morning before job interview.
p99-106Gordon Moore Redwood City / 2y San Jose State, Betty Whitaker 1950 UCB / CalTech / JHU telemetry / applied to GE, hired by Schockley
p108 Moore "negative entrepreneur", escaping prior unhappy job
p111 Rice and chessboard story, "halfway down the board, he has already committed to all the rice in China". Perhaps loose with "halfway", if 50K grains per kilogram, 200kg per million grains, then 2^33 grains (squares 1 to 32) is 170 metric tonnes. Current production is 200 million tonnes, 20 more squares; historically, production might be 1/16th of that, so 16 more squares and 2 more rows. I'd call that 3/4 of the way down the board. The exponential analogy is apt, but the description is sloppy. The author (a reporter) should test his statements (reporters rarely do).
p112 "memory chips capable of storing more than one trillion bytes" ... nope. In 2014, a dense memory process might be 20nm linewidth, and a 4 level/2 bit memory cell might be 40 by 40 nanometers. One trillion bytes is 4 trillion 2 bit memory cells, 2 million cells on a side (not counting periphery and driver cells). 2 million times 40 nanometers would be a chip more than 8 centimeters on a side ... probably a lot more. Again, the reporter/author should learn to use a calculator. No citations; POOMA?
From wikipedia: In 2019, Samsung produced a 1 TB flash chip with eight stacked 96-layer V-NAND dies, along with quad-level cell (QLC) technology (4-bit per transistor), equivalent to 2 trillion transistors, the highest transistor count of any IC chip that's a terabyte, but 8 stacked chips. This article mentions 16 layers, not 96. As of 2020/03, Samsung has just introduced a 512GB product.
Intel
p133 1969 Intel first product 3101 64 bit Schottky TTL SRAM, 2x faster than Fairchild's
"didn't leak power" ?huh? 525 mW datasheet, 16 pins, 4 bits, 4 address, 35 to 50ns
recent teardown, also 74S289
p136 1969 1101, 256 bit static PMOS silicon gate SRAM datasheet 8 address din/dout 16 pin 1.5us
- October 1970 1103 DRAM, 18 pin 10 address 16V 300ns access time 580 ns cycle time 9um PMOS
p148 Busicom, Marcian "Ted" Hoff 1937
p150 12 chips 3K-5K transistors each
p153 Noyce told Hoff go ahead, Grove not told,risking Intel
p158 Hoff on road pleasing customers, Masatoshi Shima architect, Frederico Faggin helps finish 4004
p159 Faggin bootstrap gate
p160 4001 2K ROM (Stan Mazor) 4002 320 bit RAM 4003 10 bit I/O register 4004 CPU
p162 Feb 1971 Intel 4004 16 pin 740MHz 4 bit 10um 2250 transistors
p166 1103 destructive read bug resembled core memory problem
p169 Microsystems International Canada (MIL of Bell Canada) second source
p177 1201 for Computer Terminal Corporation (CTC, now Datapoint) becomes 8008 - 800KHz
p184 8008 announced April 1972, Intellec MIC-8 design tools sales take off
p192 Bill Davidow marketing, Dov Frohman invents EPROM
p200 1979 Intel 8086 bested by Motorola 68000
p205 Peter Drucker "Nobody pays for a 'product'. What is paid for is satisfaction"
p207 Project Crush - sell customers complete environment, not just best CPU
p211 IBM young Turk Don Estridge +5 engineers to skunkworks at Boca Raton
p219 8051 (1980) cloned by 20 chip companies, "single most popular product in history"
p221 Meritocracy, executive berated wife to bring cheap car, not the Mercedes
p225 In 1990s, estimated every dollar of founder's stock returned $270,000 (1968 to 1995 59% growth per year)
- 1971/10/13 IPO price 23.50 per share, 350,000 shares, 2 cents after 13 subsequent stock splits, $50.08 on 2020/09/04, mkt cap 213B
p230 1969, Noyce and 28yo mask designer Barbara Maness employee 43, divorced Betty 1974
p239 1972-1975 Gordon Moore's "$15M" Microma watch
p247 Faggin started Zilog 1974, nonperson at Intel until 2009 and documentary The Real Revolutionaries (not available)
p251 In 1975 Wozniak used a $20 6502, couldn't afford an 8080 because Jobs withheld $3K from an Atari consulting contract
p268 Grove memo to Moore 'Intel should contract with you not to go on vacation - the world invariably seem to turn into a piece of shit while you're away."
p270 iAPX 432 fail then 960MX, and BiiN failures -> i960, not mentioned
p279 1974 November Noyce marries Ann Bowers
p282 HP's Ed Hayes quality graph, Japanese >90%, US 60% to 70%
p283 para 3 last sentence "Moore's haw" ... ha ha
p290-p348 Andy Grove biography, mention's his book Swimming Across
p344 Wrote "Physics and Technology of Semiconductor Devices" in 1967 while at Fairchild
p345 1996 "Only The Paranoid Survive", strategic inflection point: radical shift or die.
p351 gray market, discarded chips resold, organized crime and theft of scrap
p355 1974 incomplete Aloha 4 fab guarded by Doberman "Aloha Doghouse"
p356 October 1981 to March 1982; 125 Percent Solution - two extra hours without pay, 125% solution sweatbands
p358 March 1982: 80186 and 80188 microcontrollers, 82586 coprocessor, and 80286
p370 Craig Barrett Copy Exactly 1985, Fab 5 (Aloha) technological development
p372 yield to 80%, equipment utilization to 60%
p378 Noyce to SIA
p390 1985 Grove and Moore direct Intel to exit memory business, cut 7200 jobs (1/3) and closed 7 factories
p393 1985 80386 John H. Crawford
p396 Intel vs AMD, then NEC
p399 Noyce mentors Steve Jobs
p402 1987 Sematech, $100M for 5 years, TJ Rodgers (Cypress) disparages
p406 1988 Intel improves Toshiba's flash memory, two decades until Samsung outcompetes
p412 1992 June 2 death of Bob Noyce
p419 1999 April i486/80486 $950 50MHz, flash $500M business by mid-1990s
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p