Steve Blank The Secret History of Minnesota Part 1: Engineering Research Associates

Steve Blank The Secret History of Minnesota Part 1: Engineering Research Associates

This post is the latest in the “Secret History Series.” They’ll make much more sense if you watch the video or read some of the earlier posts for context. See the Secret History bibliography for sources and supplemental reading.


No Knowledge of Computers

Silicon Valley emerged from work in World War II led by Stanford professor Fred Terman developing microwave and electronics for Electronic Warfare systems. In the 1950’s and 1960’s, spurred on by Terman, Silicon Valley was selling microwave components and systems to the Defense Department, and the first fledging chip companies (Shockley, Fairchild, National, Rheem, Signetics…) were in their infancy. But there were no computer companies. Silicon Valley wouldn’t have a computer company until 1966 when Hewlett Packard shipped the HP 2116 minicomputer.

Meanwhile the biggest and fastest scientific computer companies were in Minnesota. And by 1966 they had been delivering computers for 16 years.

Minneapolis/St. Paul area companies ERA, Control Data and Cray would dominate the world of scientific computing and be an innovation cluster for computing until the mid-1980s. And then they were gone.

Why?

Just as Silicon Valley’s roots can be traced to innovation in World War II so can Minneapolis/St. Paul’s. The story starts with a company you probably never heard of – Engineering Research Associates.

It Started With Code Breaking
For thousands of years, every nation has tried to keep its diplomatic and military communications secret. They do that by encrypting (protecting the information by using a cipher/code) to scramble the messages. Other nations try to read those messages by attempting to break those codes.

During the 1930s the U.S. Army and Navy each had their own small code breaking groups. The Navy’s was called CSAW (Communications Supplemental Activity Washington) also known as OPS-20-G. The Army codebreaking group was the Signal Intelligence Service (SIS) at Arlington Hall.

The Army focused on decrypting (breaking/decoding) Japan’s diplomatic and Army codes while the Navy worked on breaking Japan’s Naval codes. This was not a harmonious arrangement. The competition between the Army and Navy code breaking groups was so contentious that in 1940 they agreed that the Army would decode and translate Japanese diplomatic code on the even days of the month and the Navy would decode and translate the messages on the odd days of the month. This arrangement lasted until Dec. 7, 1941.

At the start of WWII the Army and Navy code breaking groups each had few hundred people mainly focused on breaking Japanese codes. By the end of WWII, with the U.S. now fighting Germany, and the Soviet Union looming as a potential adversary U.S. code breaking would grow to 20,000 people working on breaking the codes of Germany, Japan and the Soviet Union.

The two groups would merge in 1949 as the Armed Forces Security Agency and then become the National Security Agency (NSA) in 1952.

The Rise of the Machines in Cryptography
Prior to 1932 practically all code breaking by the Army and Navy was done by hand. That year they began using commercial mechanical accounting equipment – the IBM keypunch, card sorters, reproducers and tabulators. The Army and Navy each had their own approach to automating cryptography. The Navy had a Rapid Analytical Machines project with hopes to build machines to integrate optics, microfilm and electronics into cryptanalytic tools. (Vannevar Bush at MIT was trying to build one for the Navy.) As WWII loomed, the advanced Rapid Machines projects were put on hold, and the Army and Navy used hundreds of specially modified commercial IBM electromechanical systems to decrypt codes.

Read the sidebars for more detailed information

Electromechanical Cryptologic Systems in WWII

By the spring 1941, the Army built the first special-purpose cryptologic attachment to the IBM punched card equipment – the GeeWhizzer using relays and rotary switches to help break the Japanese diplomatic codes. That same year, the Navy received the first in a series of 13 electro-mechanical IBM Navy Change Machines to automate decrypting cipher systems used by the Japanese Navy. The Navy attachments were extensive modifications of IBM’s standard card sorters, reproducers and tabulators. Some could be manually reconfigured via plugboards to do different tasks.

During the war the Army and Navy built ~75 of these electro-mechanical and optical systems. Some were standalone units the size of a room.

However, the bulk of the cryptoanalysis was done with IBM punch cards, sorters and tabulators, along with special microfilm comparators from Eastman Kodak. By the end of the War the Army and Navy had 750 IBM machines using several million punch cards every day.

IBM’s other mechanical contribution to cryptanalysts was the Letterwriter, (codenamed CXCO) a desktop machine that tied together electric typewriters to teletype, automatic tape and card punches, microfilm and eventually to film-processing machines. By adding plug-boards they could automate some analysis steps. Hundreds of these were bought.

The Navy’s most advanced cryptographic machine work in WWII was building 125 U.S. versions of the British code breaking machine called the BOMBE. These electromechanical BOMBES were used to crack the ENIGMA, the cipher machine used by the Germans.

Designed by the Navy’s OPS-20-G team and built at National Cash Register (NCR) in Dayton, this same Computing Machine Lab would build ~25 other types of electromechanical and optical machines, some the size of a room with 3,500 tubes, to assist in breaking Japanese and German codes. By the end of the war the Naval Computing Machine Lab was arguably building the most sophisticated electronic machines in the U.S. However, none of these machines were computers. They had no memory, and both were “‘hard-wired” to perform just one task.

(Meanwhile in England the British code breaking group in Bletchley Park built Colossus, arguably the first digital computer. At the end of the War the British offered the Navy OPS-20-G code breaking group a Colossus but the Navy turned it down.)

Dual-Use Technology
As the war was winding down, the leadership of the Navy Computing Machine Lab in OPS-20-G was thinking about how they could permanently link commercial, academic and military computing science and innovation to the Navy. After discovering that no commercial company was willing to continue their wartime work of building the specialized hardware for codebreaking, the Navy realized they needed a new company. The decided that the best way to do that was to encourage a private for-profit company to spin out and build advanced crypto-computing systems.

The Secretary of the Navy gave his OK and three officers in the Navy’s code breaking group (Commander Howard Engstrom, who had been a math professor at Yale; Lieutenant Commander William “Bill” Norris, an electrical engineer; and their contracting officer Captain Ralph Meader,) agreed to start a civilian company to continue building specialized systems to help break codes. While unique for the time, this public-private partnership was in-line with the wartime experiment of Vannevar Bush’s OSRD – using civilians in universities to develop military weapons.

Why Minneapolis/St. Paul?
While it seemed like a good idea and had the Navy’s backing, the founders got turned down for funding by companies, investment bankers and everyone, until they talked to John Parker.

Serendipity came to Minneapolis-St. Paul when the Navy team met John Parker. Parker was a ex Naval Academy graduate and a Minneapolis businessman who owned a glider manufacturing company and was well connected in Washington. Parker agreed to invest. In January 1946, they founded Engineering Research Associates (ERA). Parker became President, and got 50% of the company’s equity for a $20,000 investment (equal to $315K today) and guaranteed a $200,000 line of credit (equal to $3M today). The professional staff owned the other 50%. The new company moved into Parker’s glider hanger. Norris became the VP of Engineering, Engstrom the VP of Research, and Meader VP of Manufacturing.

The company hit the ground running. 41 of the best and brightest ex-Navy technical team members of the Naval Computing Machine Lab in Dayton moved and became the initial technical staff of ERA. When the Navy added their own staff from the Dayton Laboratory the ERA facility was designated a Naval Reserve Base and armed guards were posted at the entrance. The company took on any engineering work that came their way but were kept in business developing new code-breaking machines for the Navy. Most of the machines were custom-built to crack a specific code, and increasingly used a new ERA invention – the magnetic drum memory to process and analyze the coded texts.

ERA’s headcount grew rapidly. Within a year the company had 145 people. A year later, 420. And by 1949, 652 employees and by 1955, 1400.  Sales in their first fiscal year were $1.5 million ($22 million in today’s dollars).

During World War II the demands of war industries caused millions more Americans to move to where most defense plants located. Post-war era Americans were equally mobile, willing to move where the opportunities were. And if you were an engineer who wanted to work on the cutting edge of electronics, and electromechanical systems, ERA in Minneapolis-St. Paul was the place to be. (Applicants were told that ERA was doing electronics work for government and industry. Those who wanted more detail were given a number of cover stories. Many were told that ERA was working on airline seat reservation systems.)

How Did ERA Grow So Quickly?
The Navy thought of ERA as its “captive corporation.” From the first day ERA started with contracts from the Navy OPS-20-G codebreaking group. ERA built the most advanced electronic systems of the time. Unfortunately for the company they couldn’t tell anyone as their customer was the most secret government agency in the country – the National Security Agency.

ERAs systems were designed to solve problems defined by their Navy code-breaking customer. They fell into two categories: some projects were designed to automate existing workflows of decoding known ciphers; others were used to discover breaks into new ciphers. And with the start of the Cold War, that meant Soviet cryptosystems. ERAs cryptanalytic devices were most often designed to break only one particular foreign cipher machine (which kept a stream of new contracts coming.) The specific purpose and target of each of these systems with colorful codenames are still classified.

What Did ERA Build For the National Security Agency (NSA)?

By the end of ERA’s first year, ERA had contracts for a digital device called Alcatraz which used thousands of vacuum tubes and relays. A contract for a system named O’Malley followed. Then two “exhaustive trial” systems called Hecate for $250,000 ($3.2 million in today’s dollars) and the follow-on system, Warlock ($500,000 – $6.4 million today.) Warlock was so large that it was kept at the ERA factory and operated as a remote operations center.

Next were the Robin machines, a photoelectric comparator, used to attack the Soviet Albatross code. The first two were delivered in the end of 1950. Thirteen more were delivered to NSA over the next two years.

ERA Disk Drives
One of the problems code breakers had was the difficulty of being able to store and operate on large sets of data. To do so, cryptanalysts used thousands of punched cards, miles of paper tapes and microfilm. ERA was the pioneer in the development of an early form of disk drives called magnetic drum memories.

ERA used these magnetic drums in the special systems they built for NSA and later in their Atlas computers. They also sold them as peripherals to other computer companies.

Goldberg, which followed, was another room-sized special purpose machine – a comparator with statistical capabilities – that took photoelectric sensing and paper tape scanning to new heights.

Costing $250,000 ($3.2 million in today’s dollars), it had 7,000 tubes and was one of the first Agency machines to use a magnetic drum to store and handle data.

Another similarly sized system, Demon, followed. It was a dictionary machine designed to crack a Soviet code. It also used 34-inch-diameter magnetic drum to perform a specialized version of table lookup. Three of these large systems were delivered.

ERA engineers operated at the same relentless and exhausting pace as they had done in war time – similar to how Silicon Valley silicon and computer companies would operate three decades later.

For the next decade ERA would continue to deliver a stream of special-purpose code breaking electronic systems and subsystems for the Navy cryptologic community. (These NSA documents give a hint at the number and variety of encryption and decryption equipment at NSA in the early 1950’s: here, here, here, here, and here.)

ERA was undercapitalized and always looking for other products to sell. At the same time ERA was building systems for the NSA they pursued other lines of businesses; research studies on liquid fueled rockets, aircraft antenna couplers (which turned into a profitable product line,) a Doppler Miss Distance Indicator, Ground Support Equipment (GSE) for airlines, and Project Boom to produce instrumentation for what would become  underground nuclear tests. A 1950 study for the Office of Naval Research called High-Speed Computing Devices – a survey of all computers then existent in the U.S. As there was no single source of information about what was happening in the rapidly growing computer field, this ERA report became the bible of early U.S. computers.

The Holy Grail – A Digital Computer for Cryptography?
As complicated as the ERA machines were, they were still single function machines, not general purpose computers. But up until 1946 no one had built a general purpose computer.

With the war over what the Navy OP-20-G’s and Army SIS computing wizards really wanted was to create a single machine that could perform all the major cryptanalytic functions. The most important of the crypto techniques were based upon either locating repeated patterns, tallying massive numbers of letter patterns, and recognizing plain text, or performing some form of “exhaustive searching.”

How the NSA Got Their First Computers

Their idea was to put each of these major cryptanalytic functions in separate, dedicated, single-function hardware boxes and connect them through a central switching mechanism. That would allow cryptanalysts to tie them together in any configuration; and hook it all to free-standing input/output mechanisms. With a stock of these specialized boxes the agencies believed they could create any desired cryptanalytic engine.

Just as the consensus for this type of architecture was coalescing, a new idea emerged in 1946 – the concept of a general purpose digital computer with a von Neumann architecture. In contrast to having many separate hardwired functions, a general purpose computer would have just the four basic arithmetic ones (add, subtract, multiple and divide) along with a few that allowed movement of data between the input-output components, memory, and a single central processor. In theory, one piece of hardware could be made to imitate any machine through an inexpensive and easily changed set of instructions.

Opponents to the project believed that a von Neumann design would always be too slow because it had only a single processor to do everything. (This debate between dedicated special purpose hardware versus general purpose computers continues to this day.)

The tipping point in this debate happened in 1946 when an OPS-20-G engineer went to the Moore School’s 1946 summer course on computers. The Moore School’s computer group had just completed the ENIAC, arguably the first programmable digital computer, and they were beginning to sketch the outlines of their own new computer, the UNIVAC the first computer for business applications. The engineer came back to the Navy computing group an advocate for building a general-purpose digital computer for codebreaking having convinced himself that most cryptanalysis could be performed through digital methods. He prepared a report to show that his device would be useful to everyone at OP-20-G. The report remained Top Secret for decades.

The report detailed how a general-purpose machine could have successfully attacked the Japanese Purple codes as well as German Enigma, and Fish systems, and how it would be usefully against the current Soviet and Hagelin systems.

This changed everything for the NSA. They were now in the computer business.

ERA’s ATLAS
In 1948 the Navy gave ERA the contract to produce its first digital computer called ATLAS to be used by OPS-20-G for codebreaking.

Twenty four months later, ERA delivered the first of two 24-bit ATLAS I computers. The Atlas was 45’ wide and 9’ long. It weighed 16,000 pounds and was water cooled. Each ATLAS I cost the NSA $1.3 million ($16 million in today’s dollars).

In hindsight, the NSA crossed the Rubicon when the ATLAS I arrived. Today, an intelligence agency without computers is unimaginable. Its purchase showed incredible foresight and initiated a new era of cryptanalysis at the NSA. It was one of the handful of general purpose, binary computers anywhere. Ten years later the NSA would have 53 computers.

ERA asked the NSA for permission to offer the computer for commercial sale. The NSA required ERA to remove instructions that made the computer efficient for cryptography, and that became the commercial version – the ERA 1101 announced in December 1951. It had no operating or programming manual and its input/output facilities was a typewriter, a paper tape reader, and a paper tape punch. At the time, no programming languages existed.

ERA had delivered a breakthrough computer without having an understanding of its potential application or what a customer might have to do to use the machine. In search of commercial customers, ERA set up a ERA 1101 computer in Washington and offered it to companies as a remote computing center. As far as the commercial world knew ERA was a startup with no real computing expertise and this was their first offering. In addition, the only people with experience in writing applications for the 1101 were hidden away at NSA, and ERA was unable to staff the Arlington office to create programs for customers. Finally, ERA’s penchant for extreme secrecy left them unschooled in the art of marketing, sales, and Public Relations. When they couldn’t find any customers they donated the ERA 1101 to Georgia Tech.

With their hands on their first ever general purpose digital computer, the Navy and ERA rapidly learned what needed to be improved. ERA’s follow-on computer, the ATLAS II was a 32-bit system with additional instruction extensions for cryptography. Two were delivered to NSA between 1953 and 1954. ATLAS II cost the NSA $2.3 million ($35 million today.)

Late in 1952, a year before the ATLAS II was delivered to the NSA, ERA told Remington Rand (who now owned the company) the ATLAS II computer existed (and the government had paid for its R&D costs) and it was competitive with the newly announced IBM 701. When the ATLAS II was delivered to the NSA in 1953 they again asked for permission to sell it commercially (and again had to remove some instructions) which turned the Atlas II into the commercial ERA/Univac 1103. (see its 1956 reference manual here.)

This time with Remington Rand’s experience in sales and marketing, the computer was a commercial success with about twenty 1103s sold.

ERA’s Bogart
In 1953, with the ATLAS computers in hand, the Navy realized that a smaller digital computer could be used for data conversion and editing, and to “clean up” raw data for input to larger computers. This was the Bogart.

Physically Bogart was a “small, compact” (compared to the ATLAS) computer that weighed 3,000 pounds and covered 20 square feet of floor space. To get a feel of how insanely difficult it was to program a 1950’s computer take a look at the 1957 Bogart programming manual here.) The Bogart design team was headed by Seymour Cray. ERA delivered five Bogart machines to NSA.

Seymour Cray would reuse features of the Bogart logic design when he designed the Navy Tactical Data System computers, the UNIVAC 490 and the Control Data Corporation’s CDC 1604 and CDC 160.

By 1953, 40% of the University of Minnesota electrical engineering graduates – including Cray –  were working for ERA.

The End of an ERA
By 1952, the mainframe computer industry was beginning to take shape with office machine and electronics companies such as Remington Rand, Burroughs, National Cash Register, Raytheon, RCA and IBM. Parker, still the CEO, realized that the frantic chase of government contracts was unsustainable. (The relationship with the NSA’s procurement offices now run by Army staff, had become so strained that the Navy Computing Lab was unable to get an official letter of thanks sent to ERA for having developed the ATLAS.)

Parker calculated that ERA needed $5 million to $10 million ($75 to $150 million in today’s dollars) to grow and compete with the existing companies in the commercial computing market. Even after the NSA took over the cryptologic work of OPS-20-G the formal contracts with ERA were done through the Navy’s Bureau of Ships. NSA was known as No Such Agency and on paper its relationship with ERA didn’t exist. As far as the public knew, ERA’s products were for “the Navy.” Given that ERA’s extraordinary technical work was unknown to anyone other than the NSA, Parker didn’t think he could raise the money via a public offering (venture capital as we know it didn’t exist.)

Instead, in 1952, Parker sold ERA to Remington Rand (best known for producing typewriters) for $1.7M (about $12M in today’s dollars.) A year earlier, Remington Rand had bought Eckert-Mauchly – one of the first U.S. commercial computer companies – and its line of UNIVAC computers. They wanted ERA to get its government customers. ERA remained a standalone division. The ERA 1101 and 1103 became a part of the UNIVAC product line.

Parker became head of sales of the merged computer division. He left in 1956 and years later he became chairman of the Teleregister Corporation, the predecessor to Bunker-Ramo. He went on to become a director of several companies, including Northwest Airlines and Martin Marietta.

Remington Rand itself would be acquired by Sperry in 1955 and both ERA and Eckert–Mauchly were folded into a computer division called Sperry-UNIVAC. Much of ERA’s work was dropped, while their drum technology was used in newer UNIVAC machines. In 1986 Sperry merged with Burroughs to form Unisys.

Epilogue
For the next 60 years the NSA would have the largest collection of commercial computers and computing horsepower in the world. They would continue to supplement those with dedicated special purpose hardware.

The reorganization of American Signals Intelligence, leading to the creation of the Armed Forces Signals Agency (AFSA) in 1949, then the NSA in 1952, contributed to the demise of the special relationship between ERA and the code- breakers. The integration of the Army and Navy brought a shift in who made decisions about computer purchasing. NSA inherited a computer staff from the Army side of technical SIGINT. They had different ties and orientations than the few remaining old Navy hands. As a result, the new core NSA group did not protest when the special group that integrated Agency and ERA work was disbanded. The 1954 termination of the Navy Computing Machine Lab in St. Paul went almost unnoticed.

But the era of Minnesota’s role as a scientific computing and innovation cluster wasn’t over. In fact, it was just getting started. In 1957 ERA co-founder William Norris, and Sperry-Univac engineers Seymour Cray, Willis Drake, and ERA’s treasurer Arnold Ryden, along with a half dozen others, left Sperry-Univac and teamed up with three investors to form a new Minneapolis-based computer company: Control Data Corporation (CDC). For the next two decades Control Data would build the fastest scientific computers in the world.

Read all the Secret History posts here


Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *