# Edson de Castro

# Edson de Castro

Edson Donald de Castro (September 14, 1938 – September 6, 2024) was an American electrical engineer and business executive who worked on [Digital Equipment Corporation](/organizations/digital-equipment-corporation)'s PDP-8 and co-founded the minicomputer company Data General. At DEC he worked with Gordon Bell on the PDP-5 and then designed the first version of the PDP-8; after DEC declined a family of 16-bit computers he wanted to build, he and other engineers started Data General in 1968. The company's first machine, the 16-bit Data General Nova, sold 500 units in the 15 months after the first shipment, each for less than $10,000. He led Data General for roughly two decades. [Tracy Kidder](/people/tracy-kidder)'s The Soul of a New Machine (1981), an account of the company's effort to build a 32-bit superminicomputer, discusses his management of that business.

He was born in Plainfield, New Jersey, in 1938. The family lived in Ohio while he was between about two and four, returned briefly to New Jersey, and settled in Newton, Massachusetts, when he was about six. His father, Edson E. de Castro, was a chemical engineer; de Castro was the eldest of three children and spent twelve years in parochial schools, graduating from St. Mary's High School in 1956. In 1960 he received a bachelor's degree in electrical engineering from Lowell Technological Institute, now the University of Massachusetts Lowell. He had a partial scholarship to the Massachusetts Institute of Technology, where his father had studied, and chose Lowell instead because it offered a full scholarship and he expected more contact with faculty. He served in the Air National Guard reserves from 1960 to 1968. He left DEC for the first year of the MBA program at Harvard Business School and did not continue, which he attributed to his grades.

After Air National Guard basic training in 1961 he joined DEC as an applications engineer. In the custom-products group he worked on a computer for the Canadian atomic-energy authority that became the PDP-5, then on the upgrade that became the PDP-8, and later on an integrated-circuit version, the PDP-8/I. After IBM announced the System/360 in 1964, he proposed that DEC replace its 12-, 18-, and 36-bit machines with one compatible family, called the PDP-X inside the company; the proposal was turned down. Planning for a new firm began in mid-1967 with DEC colleagues Henry Burkhardt III and Richard Sogge and, from Fairchild Semiconductor, salesman Herbert Richman. They began full-time work on April 15, 1968, from a rented storefront in Hudson, Massachusetts. De Castro was 29. Lawyer Frederick Adler put together a financing of $800,000 in two tranches, with George Cogar, a founder of Mohawk Data Sciences, as the lead investor. De Castro did most of the Nova's logic design: a machine that looked like a 16-bit computer to the programmer and was built four bits at a time, on large printed-circuit boards, to match the integrated circuits then sold. The company went public in November 1969.

In a 2002–2003 oral history, de Castro said he chartered Tom West's Eagle project—the 32-bit Eclipse-compatible machine Kidder followed—because a more ambitious architecture, the Fountainhead project, was not going to ship in time. He later argued that the proprietary line had run its course and put development money into Unix machines based on a Motorola microprocessor, the AViiON line, and into a fault-tolerant disk system called CLARiiON. He said the board wanted the company to stay with its existing business and moved to replace him. In August 1999, with Ronald Skates as Data General's chief executive, EMC Corporation agreed to buy the company in a stock swap valued at $1.1 billion, describing midrange storage as the point of the deal. After leaving Data General, de Castro ran a small biotechnology company in Boulder, Colorado, for about a year and a half; the firm used gene-expression methods to test whether compounds were toxic and was later sold. He also served on boards, including those of Boston University and Brigham and Women's Hospital, and was a founder of the Massachusetts High Technology Council. He lived in Boylston, flew airplanes and helicopters, and died on September 6, 2024, eight days before his 86th birthday. In 2026 the university named a renovated Ball Hall lecture hall for him and created an engineering professorship funded by a family bequest.

## Early life and education

### Family and schooling

Edson Donald de Castro was born on 14 September 1938 in Plainfield, New Jersey, to Barbara (Butt) de Castro and Edson E. de Castro, a chemical engineer and graduate of the Massachusetts Institute of Technology. The family moved with his father’s projects: from about age two to four they lived in Ohio, where his father helped design processes for a [Procter & Gamble](/organizations/procter-and-gamble) plant, then returned to New Jersey, and at six settled in Newton, Massachusetts.

He was the eldest of three. A sister, Sally, two years younger, earned a physics degree and later a doctorate and by 2002 was superintendent of schools in Watertown, Massachusetts; a brother, John, eight years younger, became a psychology professor at Georgia State University. De Castro, who did not take a doctorate, later called himself the least formally educated of the three. He spent twelve years in parochial schools and in 1956 graduated from St. Mary’s High School, a Christian Brothers college-preparatory school whose fixed course included four years of Latin, four years of mathematics, and a year each of the basic sciences, plus history and English. He took apart bicycles and model trains and, as a teenager, repaired television sets, work that paid well and confirmed an interest in engineering and electronics.

### Lowell Tech, the Air National Guard, and Harvard

MIT, his father’s school, was the obvious choice, and de Castro received a partial scholarship after applying and interviewing. He enrolled instead at Lowell Technological Institute, the state engineering school in Lowell that is now the University of Massachusetts Lowell. He later said he and his father preferred the smaller school, expecting more faculty attention for undergraduates than at MIT. At a 2026 ceremony his sister said he chose Lowell over MIT because it offered a full scholarship. Freshmen shared a curriculum and declared a major only as sophomores. He recalled that admission was easy and attrition high—the sophomore class, he said, was about a third the size of the freshman class—and that the institute then produced about 2 percent of U.S. engineering graduates. He concentrated on electrical engineering. In 1960 he graduated cum laude; a senior course in pulse and digital circuits still used a vacuum-tube textbook, and he later said he had taken no circuit-design course taught in solid-state devices.

Under the Selective Service System, college deferments ended at graduation. Air Force Reserve Officers’ Training Corps was mandatory for his first two years at Lowell. He was interested in pilot training, but nearsightedness stopped that after the required physical. He joined the Air National Guard so a draft deferment would not tie him to one kind of employer. After graduation he spent about eight weeks in basic training at Lackland Air Force Base in Texas, finished by Christmas 1960 on his account, and then served in the reserve for eight years, with two weeks of duty each summer, from 1960 to 1968.

Before basic training he applied to Harvard Business School, which deferred him a year rather than take him straight from college. Around January 1961 he joined Digital Equipment Corporation in Maynard, Massachusetts, as an applications engineer. He left in late August and that September, about a year after graduating, began the two-year MBA program, commuting from Newton. The case method, with courses such as accounting (called “control”), marketing under [Theodore Levitt](/people/theodore-levitt), administrative practices, and manufacturing, demanded argument rather than a single engineering answer, and he found it difficult. After one year his grades were too low to continue, and an appeal failed. Already hired for the summer in Digital’s custom-products group, he stayed. He later called the unfinished year the most valuable of his education.

## Years at Digital Equipment Corporation

De Castro joined Digital Equipment Corporation in Maynard, Massachusetts, in about January 1961, shortly after Air National Guard basic training. The company, with about a hundred employees, sold logic modules and had built the PDP-1 to demonstrate them. Stan Olsen, who ran sales, hired him as an applications engineer—the third in that group—to help customers design with the modules. He left at the end of that summer for the first year of the MBA program at Harvard Business School. His grades were not high enough to continue, and DEC took him back into special systems.

### The PDP-5 and the PDP-8

The machine the Computer History Museum credits him with designing is the PDP-8, introduced in 1965. A DEC specification preserved by the museum lists a 12-bit word, a 1.5-microsecond cycle, 4,096 words of core memory, a Teletype, and a price of $18,000. The exhibit describes more than a dozen variations and sales of over 10,000 machines in 25 years, and treats the PDP-8 as a later version of a small stored-program computer he and Gordon Bell substituted for a hard-wired reactor monitor. It kept the instruction set of the PDP-5._timeline_1957-1997.102630354/1965-2.htm)

Bell dated a 12-bit controller study, the DC-12, drawing on the LINC, to talks with Foxboro in the fall of 1961. DEC instead sold Atomic Energy of Canada Limited an 18-bit PDP-4 for reactor monitoring at Chalk River Laboratories. That job still needed a front end to convert analog signals, check limits, and survive a PDP-4 failure. After a visit in the winter of 1962, Bell wrote, DEC chose a 12-bit design based on the DC-12 over a 10-bit LINC-line machine too small for the data and the program. He credited Alan Kotok and himself with the instruction set, and de Castro, the applications engineer on the analog front end, with the logic.

De Castro remembered widening the sketch to 12 bits to match 6-bit peripheral characters, adding subroutines, keeping an instruction for part of the analog-to-digital conversion, and allowing 4,096 words rather than 1,024. Management, he said, would not call so cheap a machine a computer, so it stayed the DC-12 until a trade show renamed it the PDP-5. He wrote the Chalk River monitor. The customer kept the quoted price, about $50,000, and took the larger memory. A 1964 DEC advertisement offered the PDP-5 at $27,000. A later count put production at 116 machines, withdrawn in early 1967.

After further PDP-5 customer work he designed the faster, cheaper PDP-8 on Flip Chip modules meant to carry hybrid integrated circuits. The hybrids never became practical. Once [Ken Olsen](/people/ken-olsen) judged them too expensive, the machine used discrete parts on smaller double-sided cards. Richard Sogge worked on the memory circuits; de Castro said he did the logic while still assigned to special systems, not engineering. Nick Mazzarese, from sales, then became product-line manager for the 12-bit machines, and de Castro worked for him._timeline_1957-1997.102630354/1965-2.htm)

Around 1967 he did the integrated-circuit PDP-8/I: monolithic circuits on Flip Chips, a wire-wrapped backplane, a faster core, and a rack drawer instead of swinging doors, with the old software unchanged. He said DEC’s logic symbols did not fit integrated circuits, so he used military-standard symbols without a new drafting rule and was criticized for it. The museum’s account of the family records the same move to integrated-circuit Flip Chip cards.

### The PDP-X proposal

The IBM System/360, announced in 1964, was a compatible family built on 8-bit bytes. Customers asked whether DEC’s incompatible 12- and 18-bit lines would follow. Scientific Data Systems was moving from 24-bit machines toward 32 bits, and Computer Control Corporation, later acquired by Honeywell, was planning a 16-bit computer. Cheaper monolithic circuits also made DEC’s small modules and wired backplanes look, to him, like needless cost.

He asked the Computer Guidance Committee—Olsen, Mazzarese, Stan Olsen, and, he thought, Bell—for a byte-oriented family implementable at 8, 16, and 32 bits, with common software, on large boards rather than gate-sized modules. Marketing opposed forcing users to rewrite software; manufacturing and field service opposed the boards. The committee rejected the proposal.

DEC archive memoranda, dated by Bob Supnik from June 1967 to February 1968, specify the PDP-X as a 16-bit multi-register design with variable-length instructions, in models sharing peripherals and software: a cheaper one against the PDP-8 and a costlier one against the PDP-9. By spring 1968 it was dropped as not a clear gain on the 12- and 18-bit lines. De Castro said he then finished the PDP-8/I, that an 8-, 16-, and 32-bit program was too large for a startup, and that the later Data General Nova was not the PDP-X architecture. Supnik’s reading against the Nova and the PDP-11 agreed: the PDP-X was a more complicated register-and-memory machine, not the prototype of either.

## Founding Data General

### Partners and capital

In mid-1967, after Digital Equipment Corporation turned down his PDP-X—a compatible 8-, 16-, and 32-bit family meant to replace DEC’s 12-, 18-, and 36-bit lines along the lines of the IBM System/360—de Castro began planning a smaller company. The group was de Castro; Pat Green, a DEC manager in custom products and special systems; Henry Burkhardt III, a programmer of about 22 or 23; and Richard Sogge, a circuit engineer. They added Herbert Richman, a [Fairchild Semiconductor](/organizations/fairchild-semiconductor) salesman who called on de Castro and Sogge. A company history gives the ages at the start, with lawyer Frederick R. Adler, as de Castro 29, Sogge 28, Burkhardt 23, Richman 32, and Adler 42; the Computer History Museum also records de Castro at 29 and Burkhardt at 23.

De Castro said the PDP-X was too large for a startup. By the end of 1967 the group had dropped an 8-bit sketch for one 16-bit machine, built internally four bits at a time to match available integrated-circuit slices, a distinction he credited to Burkhardt. That design became the Nova. He said rumors that its architecture was the PDP-X were wrong; what carried over, in his account, was the frustration. Gordon Bell, then a DEC engineering executive, later traced the spin-off to “unappreciated innovations”: the founders had worked on the PDP-8 and PDP-9, and DEC had rejected their scalable PDP-X.

They did not approach American Research and Development, which had financed DEC and which de Castro treated as conflicted, and he recalled little interest from other venture capital firms, including a Boston firm the interview transcript calls Graylock—Greylock. Richman then introduced people already successful in electronics. In de Castro’s recollection George Cogar, a founder of Mohawk Data Sciences, took about half the deal, with Seymour Schweber, an electronics distributor, and smaller investors alongside. Adler, referred through a contact of Richman’s, incorporated the company and arranged $800,000 in two $400,000 tranches. A reference history says Adler put in $50,000 and raised much of the rest; an internal history says Richman raised the $800,000 and Adler restructured it among about 80 individuals plus the five principals. De Castro, calling the figure hazy, thought outsiders received about 40 percent in that only private round, and that the founders put in cash as well. The Massachusetts Supreme Judicial Court later found that he contributed $15,000 from joint savings and, with four others, formed Data General.

### The split from DEC

Green did not resign. De Castro said that when resignation was at hand Green, about five years older and urged by his wife, feared that income and benefits such as health care would not be secure. De Castro, Burkhardt, and Sogge, with Adler’s agreement, then closed the operating role, while still letting Green invest. Richman remained part of the founding group but, by agreement, stayed off the payroll until there was something to sell, and joined once the machine could be shown.

The three engineers started full time on 15 April 1968—tax day, as de Castro remembered it—and rented a storefront in Hudson, Massachusetts. A reference history dates incorporation to that day and calls the premises a former beauty parlor. A 2001 museum account describes the Hudson building as one that was by then a beauty parlor again. De Castro took the logic, Burkhardt the software, and Sogge the memory and circuits.

He later said he had already given Ken Olsen the products that launched DEC into the minicomputer business, and that he was as free to leave as Olsen had been in leaving Lincoln Laboratory. A 1994 company history, citing an April 1979 article in *Fortune*, reports that DEC’s president said, “What they did was so bad we’re still upset about it,” and that Digital never sued for theft of intellectual property.

## The Nova and the minicomputer market

### Design

The Nova was a narrower machine than the PDP-X family Digital Equipment Corporation had rejected. Edson de Castro later said a startup needed one computer it could finish quickly, and that the Nova's design was not the PDP-X. Of the three engineers who left to build it, he did primarily the logic design, Henry Burkhardt the software, and Richard Sogge the memory and circuits.

The architecture, which he attributed to himself and Burkhardt, used four 16-bit accumulators, two of them as index registers, where most small computers had one. Arithmetic and logical instructions carried no memory address, so individual bits selected a shift, a carry treatment, and a skip; he said the format was chosen because it mapped onto hardware while parts still cost real money. Programmers saw a 16-bit computer. Inside, data moved four bits at a time, the width of the adders and registers then sold as single integrated circuits, so the machine looked slow and 16 bits wide from outside and fast and four bits wide within. In *Electronics* for December 9, 1968, the designers called those groups "nibbles": a four-bit adder finished a 16-bit addition in four steps, more than half the gates and flip-flops were medium-scale integration, and the logic family was transistor–transistor logic. They aimed at a full cycle, memory reference included, in under three microseconds. The November 1968 brochure gave the 4,096-word core module a 2.6-microsecond cycle and interchangeable read-only memory a 2.4-microsecond access time; de Castro later said the logic itself ran nearer one microsecond.

The same pressure shaped the wiring. He had argued at DEC that integrated circuits should hold the primitive functions and large boards should connect them, in place of a backplane full of wires. The Nova used double-sided boards 15 inches square: two for the processor in the first version, one for a 4,096-by-16-bit memory, and one for input-output, plugged into a printed-circuit backplane with virtually no hand wiring. *Electronics* described that memory—sixteen 64-by-64 core mats and their circuits—as a single 15-inch board about half an inch high in a frame 5¼ inches tall, and the brochure said the cabinet held seven such boards. Hadco, one of the few shops able to build boards that large, made them. Field service was meant to replace a processor or memory board, not trace a fault to a gate, a change DEC's maintainers had resisted.

### Introduction and rivals

The first working Nova went to the Fall Joint Computer Conference, delayed in 1968 from November into December. *Electronics* for December 9 treated the meeting as under way that week in San Francisco. Burkhardt programmed Spacewar! for the booth, and de Castro remembered it running there.

The November 1968 brochure, whose cover the Computer History Museum catalogues under the heading "The best small computer in the world," priced 4,096 words and a Teletype interface at $7,950. *Electronics* classed such minicomputers as machines under $10,000 and stated a quantity goal of $5,000 or less for a 4,096-word system; the museum lists serial number 1 at $9,200, with 4K words of core, in a chassis 5¼ by 19 by 21½ inches. De Castro remembered the first customer machine leaving in February 1969 for a university in Texas, lost in an airline strike, and recovered from a Chicago warehouse around May after a replacement had been sent. The museum says that serial number went to Unitech in Austin, Texas, for Mobil Oil, was lost by Braniff, and was found two months later. Early buyers, in his account, were technical university and industrial users; the starting software was an assembler and a loader, the standard peripheral a Teletype Model 33, and the brochure also listed an editor, a debugger, and floating-point routines. The museum states that 500 Novas sold in the 15 months after first shipment, at less than $10,000 each.

Other small computers were already in the market. De Castro recalled the 12-bit PDP-8, with 4K of memory, at a base price of $18,000. The PDP-8/I of 1968 kept that software, used a 1.5-microsecond memory of 4,096 12-bit words, and was priced, a DEC engineer wrote, 29 percent below the PDP-8. Sixteen-bit machines were not new either. Charles House, a former [Hewlett-Packard](/organizations/hewlett-packard) executive writing for the Computer History Museum in 2016, described the 1966 HP 2116 as an instrumentation computer and Computer Control Company's 1965 DDP-116 as an earlier 16-bit product, and noted that DEC's PDP-11 waited until 1970. A PDP-11/20 with 4K of core, a console, and an ASR-33 Teletype listed at $10,800 on November 1, 1970. De Castro later said that machine's richer instruction set, more than Hewlett-Packard's computers, was what Data General had to answer.

The Supernova kept the Nova's instruction set and changed the implementation. He described it as a fully parallel 16-bit processor for fast memory, not another four-bit data path. A brochure date-stamped February 10, 1970, specified an 800-nanosecond core cycle, 300-nanosecond read-only memory, and a basic 4K price of $11,700. He said an offer of semiconductor memory failed because the early bipolar parts ran too hot. The Nova 800 and Nova 1200, names he said indicated their memory cycle times, came next: Larry Seligman designed the faster processor and Ron Gruner the cheaper single-board one, interchangeable by a processor-board swap.

## Expansion and rivalry

### Growth and competitors

De Castro took Data General public in November 1969, about eighteen months after it started and just as it reached breakeven. Investor Fred Adler urged the sale; Bache led the underwriting, and de Castro later estimated the proceeds at about $5 million. He recalled revenue of roughly $1 million in the fiscal year ended September 1969, $7 million the next year, and $15 million the year after, with later stock offerings financing growth the company could not yet fund itself. Work moved into a building put up for the firm in Southborough, Massachusetts.

At the January 1974 annual meeting he called 1973 a year of "notable achievement": shipments had more than doubled, to more than 4,000 computers, employment had risen from 840 to 1,700, and international business was 29 percent of sales. *Computerworld* reported revenue of $53.3 million and earnings of $6.7 million, against $30.3 million and $3.9 million the year before. That month the paper noted that the shares were listed on the [New York](/organizations/new-york) Stock Exchange; a later company history dates the listing to December 28, 1973. He said the company was making memory, disk heads, and paper-tape readers itself. *Electronics* in July 1975 put fiscal 1974 sales at $83,196,000, up 56 percent, and quoted him arguing that vertical integration was necessary as mainframe makers moved down-market and microprocessor makers moved up. The magazine set the Eclipse beside Digital Equipment Corporation's PDP-11/70, Varian Associates' V-70, and Interdata's Megamini.

He treated the Eclipse as the answer to the PDP-11. Once that machine offered a richer instruction set than the original Nova, he had Larry Seligman design a still 16-bit computer, upward compatible with the Nova, able to address more memory and fitted with optional scientific or commercial instructions. Scientific and commercial models were marked S and C, and the commercial line eventually had a COBOL compiler. *Computerworld* on October 9, 1974, reported that the S/100 and S/200 had been announced in New York the week before as cheaper than the Nova 840 and, the company said, more powerful, keeping the Nova instruction set and adding others; the S/200 was offered with up to 256K bytes of memory and the S/100 with up to 64K. De Castro later said the line "competed reasonably successfully" with the PDP-11, "the major competitor," while Hewlett-Packard was present but "not as much of a factor as Digital." An IEEE account names DEC, Data General, and Prime Computer as important suppliers to finance, government, and manufacturing, and places Wang Laboratories in word processing and office systems chiefly against IBM.

Commercial work stayed at the edge of the business he ran. He said a BASIC interpreter and a FORTRAN compiler made Data General a fuller scientific and industrial supplier, "not in the commercial markets," which he was "not sure we ever really penetrated" in a meaningful way. In January 1974 he told shareholders a supermarket checkout system was still in development. He later described losing the grocery industry's barcode contest to IBM, installing systems at Shaw's, and selling the effort to Sweda because point-of-sale firms, including NCR and IBM, could reach retailers and he could not. A numerical-control product for machine tools also failed, he said, because the tool vendors already supplied the controls. A compilation of Fortune 500 rankings lists Data General 500th in 1979, with revenue of $379.9 million and profit of $40.3 million.

### Management and disputes

He remembered that Ken Olsen had predicted the company would fail, and that Digital was "a bit annoyed" when it did not. A company history reports that about ten years later Digital's president told *Fortune*, "What they did was so bad we're still upset about it," and that Digital did not sue over the departure. Reviewing Tracy Kidder's *The Soul of a New Machine* on October 8, 1981, Jeremy Bernstein quoted the book's claim that *Business Week* "blamed de Castro's style of management mainly, but appeared not to understand that style any better than anyone else," and noted that the book never identified the issue.

The lawsuits in the record concern copies and software, not that split. At a Houston trade show in the fall of 1970, John N. Ackley of Digital Computer Controls said he planned a second source for the Nova 1200; de Castro replied that Data General would sue. On November 7, 1975, the Delaware Court of Chancery found that Ackley's firm had traced confidential Nova 1200 drawings to build the substantially identical D-116, and permanently barred use of those drawings to make such a machine, leaving damages to a jury. The *[Los Angeles Times](/organizations/los-angeles-times)* reported in December 1986 that Digidyne's nine-year antitrust suit alleged Data General had tied operating-system software to its hardware; Fairchild Semiconductor, which joined in 1978, settled that September for more than $52 million, Digidyne was seeking $101 million, and the paper said Data General had been found liable earlier in 1986 for unspecified antitrust damages.

## Thirty-two-bit machines and public attention

### Fountainhead and Eagle

By the mid-1970s Data General was looking past the 16-bit Eclipse. In an oral history recorded in 2002 and 2003, Edson de Castro said the company had heard that Digital Equipment Corporation was building a 32-bit upgrade to the PDP-11 and needed one of its own. DEC president Kenneth H. Olsen announced that machine, the VAX-11/780, in October 1977; a later DEC history dates the shareholder-meeting unveiling, with VMS, to 25 October. Datapro described it as DEC’s first 32-bit architecture, an extension of the PDP-11, with about four billion bytes of virtual address space and programs of up to 32 million bytes. Basic packages were listed at $128,600 to $185,000.

De Castro said Ron Gruner’s group was asked for an advanced 32-bit architecture. It first met in a rented apartment at the Fountainhead complex, which gave the work its name, the Fountainhead Project, or FHP. He later called the design too ambitious, including segmentation that, in his view, showed up in a commercial product mainly in the IBM System/38. The group was then given its own laboratory at Research Triangle Park, North Carolina, near manufacturing Data General was already placing there. Implementation, he said, was “non-trivial,” and the project “began to go on and on” while a 32-bit machine was needed sooner.

Tom West, who had worked with Gruner on Eclipse products, proposed what de Castro called something “much less aggressive”: a 32-bit computer “Eclipse-like with extensions,” finishable in a reasonable time. West received the charter for Eagle. Data General announced it in New York in the last week of April 1980 as the Eclipse MV/8000, its first 32-bit minicomputer. Jeremy Bernstein later dated the debut to 29 April at the Roosevelt Hotel.

*Computerworld* reported Data General’s claim that the MV/8000 could run programs 16 times larger than the VAX-11/780, with 4.3 gigabytes of virtual address space, and that 16-bit and 32-bit programs could run together without a “mode bit.” Engineers on the project wrote in *Electronics* that the instruction set was a superset of the 16-bit Eclipse set, so older programs need not be recompiled, and described a pipelined, microprogrammed processor. *Computerworld* listed a basic system, without software, at $153,150 and a medium system with terminals, disk, a printer, and AOS/VS at $257,750—list prices for different configurations from the VAX packages above.

De Castro later described the choice as schedule and return. Eagle, he said, was “not as technically exciting as FHP,” but it worked, shipped, and sold; finishing Fountainhead was then cancelled as not worth the likely gain. A symmetric multiprocessing operating system was salvaged and later used in DG/UX, Data General’s UNIX. He also recalled a commercial software effort, calendaring and electronic mail on a central computer with relatively unintelligent terminals—close to later office systems, except spreadsheets. It worked inside one organization but not between organizations. He judged the company “a little early,” and the centralized design a poor fit once cheap processors put a computer on each desk. A September 1982 advertisement said Data General had entered office automation in November 1981 with the Comprehensive Electronic Office (CEO) on AOS and AOS/VS.

### The Soul of a New Machine

The wider public account was Tracy Kidder’s *The Soul of a New Machine* (Atlantic–Little, Brown, 1981), winner of the 1982 Pulitzer Prize for general nonfiction and the National Book Award for general nonfiction in hardcover.

Bernstein’s review treated the book as the story of West’s group racing the VAX, quoting Kidder’s West—“We’re gonna get schmeared if we don’t react to VAX”—and a mostly young team assembled in the fall of 1978 to finish in under a year. He called those designers the heroes. He also warned that the narrative asks readers to accept scenes it does not document: no photographs or diagrams, and a reference to *Business Week* blaming de Castro’s management style with no issue or page. Bernstein telephoned Data General; West, he wrote, said that aside from minor misquotes the book was essentially accurate. It is a shaped account endorsed in outline by its central engineer, not an unmediated record.

De Castro stays at the margin. In the review he appears chiefly as the executive whose style *Business Week*, in Kidder’s citation, is said not to have understood. Ron Gruner later wrote that Kidder avoided “management” so the book would not look sponsored, and that West and the engineering vice president wanted business decisions kept out of it. Interviewed two decades later, de Castro did not revisit those scenes. He called Eagle “extremely well chronicled” by Kidder, then drew his own line: Fountainhead was the more ambitious architecture; Eagle was the machine that could be finished and sold.

## Open systems and departure from leadership

### AViiON and the board

In oral-history interviews recorded in 2002 and 2003, de Castro said a successor to Data General’s proprietary machines had to use a high-volume processor on a single chip and a commodity operating system. He said he set senior staff to work on the premise that the past was over, because refusing the change would cost more than making it. The outcome, he said, was the AViiON, built on the Motorola 88000 and Unix, with the company’s symmetric-multiprocessing software in an operating system called DG/UX.

A February 1990 Datapro report dated the introduction to February 1989 and described DG/UX 4.1 as Data General’s Unix. It called the company one of the last vendors into RISC Unix workstations and servers, against [Sun Microsystems](/organizations/sun-microsystems), Hewlett-Packard, and Digital Equipment Corporation, while the firm also planned to keep extending its proprietary systems. For the year ended September 1989 the same report gave revenue of $1.314 billion, down 3.7 percent from $1.365 billion, and a net loss of $119.7 million after a $15.5 million loss the year before.

De Castro said sales of the proprietary MV machines, and profits, were already falling, and that the AViiON effort—mostly software—would mean two or three years without profits. In his account the board said the company had to “stick to its knitting” and could not abandon its past; he continued anyway. A fault-tolerant Unix version was to include a disk subsystem called CLARiiON. He recalled a director from the insurance industry calling that project a “dumb idea,” and said the board grew “more and more disgusted” with the cost of leaving the old line and decided it needed someone else to restore profitability.

### Removal and the sale of the company

Ronald L. Skates, a former Price Waterhouse accountant who joined in 1986, became president and chief executive in 1989, when de Castro was named chairman and chief scientist. *Computerworld* reported on 17 December 1990 that de Castro would leave on 31 December and would not stand for re-election. He said there was “a certain element of disappointment,” referring to speculation that the board had somewhat forced the decision. The magazine said insiders treated the chairmanship as a figurehead role, that authority had already been shifting to Skates, and that vice chairman Herbert J. Richman also would not stand for re-election. The same issue said about $100 million in AViiON sales for 1990 would not carry a business of roughly $1.3 billion, and quoted an open-systems marketing executive who said Data General could not match Sun, Hewlett-Packard, and DEC in workstations and would sell the AViiON as a multiuser server.

De Castro’s interview places the open-systems spending with him, over the board’s objection. A 1997 *Forbes* profile, drawing on Skates, presents the move as a decision he forced after taking charge and assigned to J. Thomas West, using off-the-shelf Motorola RISC chips and Unix. *Forbes* dated the AViiON to about 1991, later than Datapro’s February 1989, wrote that the 88000 never became popular, and reported that Skates later replaced it with Intel processors. The magazine also wrote that Data General lost market share and revenue from 1984 to 1990 and more than $500 million between 1985 and 1995, and that DEC and Wang Laboratories, like Data General, had stayed with proprietary processors and software while customers wanted equipment that could be mixed across vendors.

De Castro said that after he left, CLARiiON was sold as a separate disk system and succeeded, and he agreed that Data General probably would not have been sold to EMC Corporation without it. He said EMC had at first disdained the array, which later grew inside EMC, while little further thought was given to the AViiON line. On 9 August 1999 CNET reported that EMC would buy Data General in a stock swap valued at $1.1 billion. Skates, still chief executive, said the server business would continue, “but the focus for our people going forward and the focus for EMC will be the storage side.” EMC said it would run the servers separately and fold Clariion products into its own lines; a spokesman called storage the primary focus, and an analyst questioned whether the AViiON business would be kept.

## Later career and private life

### Work after Data General

After leaving Data General, de Castro did not describe founding another computer company. In oral-history interviews recorded in November 2002 and January 2003 at his home in Boylston, Massachusetts, he said he had turned to biotechnology and had run a small firm in Boulder, Colorado, for about a year and a half. Securities filings identify that firm as Xenometrix, Inc., which developed assay kits and software to measure how human, bacterial, and yeast cells responded genetically to chemical compounds. The company told investors the assays could reduce the share of compounds put through live-animal tests; de Castro later described the work as using gene expression to judge toxicity without feeding compounds to animals.

He was chairman of Xenometrix from 1992 to 1997 and was elected chief executive on May 26, 1995, after also holding that office from June to November 1992. He told a [Computer History Museum](/organizations/computer-history-museum) interviewer that he was unsure he had made the company successful and that it was ultimately sold. In May 2001 Discovery Partners International acquired Xenometrix for about $1.8 million net of cash, gaining a gene-profiling system for toxicology research. He contrasted the pace with computing: about fourteen years, he said, from first development to a marketed product, against computer-product cycles of roughly a year to eighteen months. Biological systems were less fully understood than electronics, he added, and drug regulation slowed the work in a way the computer industry had not faced. He said he was not a critic of the Food and Drug Administration, which had a difficult task because compounds given to people needed research to show they were safe.

He remained a director of other companies. In the same interviews he said he had stayed on boards in the computer industry through the usual round of startups and sales. A 1996 Xenometrix filing already listed him as a trustee of Boston University. A 2003 HealthGate Data filing—he had been a HealthGate director since December 1994, and a self-employed consultant since 1997—also listed him as a director of AVAX Technologies, a biopharmaceutical company, and of VCampus, a publisher of internet training courseware; as a member of the corporation of Partners Healthcare System; and as a member of the advisory council of the Clark University Graduate School of Management. The United Kingdom register of the Trustees of Boston University recorded his resignation effective September 1, 2006. His family later wrote that his boards had included Boston University and Brigham and Women's Hospital.

The Massachusetts High Technology Council dated from his Data General years, not from retirement. In a 1984 interview, council director Howard Foley named Ed de Castro, with Ray Stata, Dee D'Arbeloff, and Dean Morton, among those involved when the group was incorporated, then with 39 original members. Later filings still described him as a founder and an executive-committee member of the Massachusetts High Tech Council.

### Family and private interests

He continued to live in Boylston. In 1992 he married Eileen Riggs, who had worked at Data General, including in corporate sales; she died on April 13, 2024. He was survived by five children—Susan, Stephen, Michael, David, and Thomas—while her obituary named David and Thomas as their sons.

His family described him as an airplane and helicopter pilot who flew to places including Martha's Vineyard, the Bahamas, and Lake Powell, and who spent time maintaining airplanes, an antique boat, and antique cars. They also wrote that he preferred routes away from ordinary tourist stops, naming Yemen, Russia, Colombia, Haiti, and Guatemala among the places he liked.

## Death and recognition

### Death

Edson Donald de Castro died on 6 September 2024, aged 85, eight days before his 86th birthday. Chiampa Funeral Home in Shrewsbury, Massachusetts, recorded his birth date as 14 September 1938 and his death date as 6 September 2024, and published the notice on Legacy.com on 11 September 2024. Visiting hours were held at the funeral home, 5 Church Road in Shrewsbury Center, from 10:00 a.m. to 11:00 a.m. on 5 October 2024, followed by a funeral service; interment with military honors was at Pine Grove Cemetery in Boylston, Massachusetts. The notice recorded his service in the Air National Guard reserves from 1960 to 1968.

He was preceded in death by his wife, Eileen Riggs de Castro. Survivors named in the notice were his children Susan, Stephen, Michael, David, and Thomas, eight grandchildren, his sister Sally, and his brother John.

### Honors and historical place

In April 2026 the University of Massachusetts Lowell dedicated Ball Hall Room 214, a renovated lecture hall in the Francis College of Engineering, to de Castro and, from a family bequest, established the Edson deCastro ’60, ’78 (H) Professorship in that college. The university’s account, published on 22 April 2026 and carried by the *Lowell Sun* on 13 May, described a ribbon-cutting attended by his children, grandchildren, and siblings, at which his daughter Susan cut the ribbon.

His sister Sally, who attended, said he had chosen Lowell Tech over the Massachusetts Institute of Technology because he could get a full scholarship, and that the opportunity had made a difference. Chancellor Julie Chen told the gathering that he had started a company at age 29 and that, as the university presented it, the foundation for that work had begun on the campus.

The Computer History Museum states that de Castro worked closely with Gordon Bell on the PDP-5 at Digital Equipment Corporation and then designed the initial version of the PDP-8. In that account, DEC’s refusal to approve a family of 16-bit computers he wanted to design was followed by his departure, with other DEC engineers, to launch Data General in 1968 and to create their own 16-bit design. The university identified him as a 1960 electrical-engineering graduate of Lowell Technological Institute, a predecessor of UMass Lowell, who co-founded Data General in 1968 and introduced the Nova, which it described as one of the first 16-bit minicomputers, and said the company grew into a Fortune 500 firm.

## Dated facts

_No published facts on this page yet._

## Methodology

https://factrail.ai/methodology

Canonical: https://factrail.ai/people/edson-de-castro
