The internal history of the biggest mistake Texas Instruments, the microprocessor TMS9900 / SurprizingFacts


If you use a desktop or laptop, it probably has a microprocessor in the Intel 808x series, no matter if the machine is running Running Windows or Mac. The absolute dominance of these microprocessors goes back to 1978, when IBM chose 8088 for its first personal computer. But this choice was not at all obvious. Some believe that the Intel 8088 was the worst choice of the then available 16-bit microprocessors.

No, it was not the worst choice. He had a serious, but worst, alternative. I know this because I was in charge of the organization that worked with TI, who developed it: TMS9900. Although this stuff was used in the world's first 16-bit home computer, you probably did not hear about it. As the saying goes, the history is written by the winners.

This chapter of history is interesting not only for the chip from TI, but for its other competitor, Motorola 68000, from the technological point of view superior to both Intel 8088 and TMS9900. And yet, 68000 did not get into the IBM PC. And here's a little-known story from well-informed sources about how IBM chose an underdeveloped chip, TI spawned a loser, and the clear leader from Motorola lost.

In 1978 Year, the author held a presentation of the TMS9900 chip, which was considered as a candidate for a personal computer from IBM. TI did not get a contract.

I got a job at TI in 1972, right after the graduate school, and two years later I made a presentation for Jack Carsten, MOS department manager in Houston, Where the company's work was based on metal-oxide-semiconductor (MOS) chips. I, a young engineer, were somewhat timid before Jack, the whole presentation was sitting with legs tucked over the table in the conference room, smoking a cigar and muttering "garbage" every time he did not agree with something that I say.

At that time, a large three of the semiconductor companies – Fairchild, Motorola, and TI – were trying to move from bipolar chips to MOS. MOS chips required a completely different scheme and manufacturing process, and startups like Intel developed much faster than existing companies. Of all the big three TIs, the transition was the best, thanks in large part to people like LJ Sevin, who left TI in 1969 to found Mostek and subsequently engage in venture capital investments. Karsten, who previously worked for TI as the chief manager of the lucrative line of transistor-transistor logic (TTL) products, was also a key person in the transition to MOS.

The TOS MOS Division achieved the most significant success in developing logical chips for fast-growing Market of portable calculators. Although the company competed, and eventually won Intel, in the development of the first general-purpose processor, TI engineers were not particularly interested in Intel's microprocessors (MPs), 4-bit 4004 and 8-bit 8008. TI noticed 8-bit microprocessors 8080 and the following 8080A, because they were more promising than 4004. The MOS department was instructed to catch up with Intel on both MP and DRAM (dynamic RAM with random access, capable of cramming more memory cells onto the chip than RAM, but requiring constant updates to prevent data loss ).

So in TI and There was a strategy for creating a general purpose MP. The key assumptions of the strategy were: application software will be the main incentive for the evolution of these chips, and as the owner of the successful MOS IC line, TI will be in an excellent position to develop an industry standard for microcomputers, security systems and consumer products. All of these industries were rapidly growing sources of profit for the company. But for this TI it was necessary to jump over the current advanced 8-bit development, the example of which was Intel 8080, and to become the first company to bring to the market 16-bit architecture. This strategy spawned the plan for creating the TMS9900.

Winner: The Intel 8088 microprocessors were far from ideal, but IBM chose them for their personal computer launched in 1981.

TI ​​has already demonstrated its computer skills in the supercomputer race in the late 1960s. The oil companies moved this race, trying to achieve a competitive advantage in the three-dimensional seismic analysis while exploring oil fields. In this business, TI was founded. IBM, Control Data Corporation and other companies competed in this race, but TI first brought its Advanced Scientific Computer to the market.

So for TI, the choice of architecture for a 16-bit chip was simple. TI had a strategy of "one company, one computer architecture", aimed at using the synergistic effects that exist between different, disparate departments of the company. The Information Systems Branch has already launched a family of TTL-based microcomputers for use in the Ramada Inn hotel chain throughout the United States. So the TMS9900 had to use an architecture very similar to the architecture of the TI microcomputer.

Karsten's team knew that the development of the TMS9900 – like its bipolar version for the military called the SBP9900 – would take time, and the chips, Will not be ready before 1975-1976. At this time, the MOS branch had to act. They planned to start copying the Intel 8080A to bring something to the market, then develop an original 8-bit microprocessor architecture (which will be called the TMS5500), and finally go to the 16-bit TMS9900. National Semiconductor by that time had already released a 16-bit set of general-purpose logic chips called IMP-16, but due to the fact that it was a set of several chips, it won no popularity.

The TMS9900 had His baggage of difficulties in the development and delays, but in the end he was ready by 1976. And still he faced several major problems. First, there were no compatible 16-bit peripheral chips. And without the peripheral chips working on the transmission and storage of data, the MP will be useless for system development. The second problem was that the architecture of the 9900, similar to the one TI used in microcomputers, had only 16 bits of logical address space-just like the 8-bit processors of that time. This problem could not be solved without developing a new architecture from scratch. The last problem was that although TI could use one MP technology for its minicomputer, defense industry, and semiconductor business, its competitors in these industries would be at a disadvantage if they took the architecture from TI for their products

The loser: Among the major problems that plagued the TMS9900 was the lack of 16-bit peripheral chips, which made the MP useless in system development

To combat the lack of a 16-bit TI engineers came up with the following innovation: Yu. Why not adjust the 8-bit port on the TMS9900, so that a large number of existing peripheral chips designed for 8-bit MPs can work with it? I'm sure that at that time it seemed reasonable. As a result, in 1977 appeared TMS9980. Attaching the 8-bit periphery to the 16-bit MP eliminated the only real advantage of the 16-bit architecture: performance. The 9980 required two instruction loops to execute the command from the 8-bit periphery, resulting in an efficient performance that dropped by half, which was no better than the existing 8-bit MPs. Prior to the implementation of the great TI plan, Carsten left the company as vice president of sales and marketing at Intel, feeling beyond any doubt that Intel would become a very serious rival, which it would be extremely difficult to win in the market of the MT.

Intel , Naturally developed its own 16-bit microprocessor, the 8086, which was released in April 1978. The company approached the issue of the lack of a compatible 16-bit periphery in exactly the same way, adding an 8-bit port to the MP, which gave rise to Intel 8088. Like TI 9980, Intel 8088 was junk, and showed a reduced Speed ​​in comparison with 8086 in any real system. But the chip from Intel had one fundamental advantage over TI: 20 bits of logical address space instead of 16. And as a result, he had the opportunity to access a megabyte of memory, rather than 64K, like the TI 9900. In addition, the external registers TMS9900 and 9980

And while Intel successfully developed alternative sources of production for the 8086, TI struggled to make similar deals. At that time, most customers needed at least two competing suppliers for each new family of semiconductor components to ensure the availability of the product at an affordable price.

And at this time, several competitors announced their plans to create 16-bit Processors of general purpose. The most ambitious plan was the Motorola 68000. Although it had 16 external contacts, its internal architecture was 32-bit, and it had 24 bits for the logical address space. The next product might be able to use 32 address bits. Zilog, the creator of the popular 8-bit MP Z80, announced the development of a 16-bit Z8000, with segmented memory, the output of which was planned for 1978-1979. In contrast to the 68000, the Z8000 had a simple 16-bit architecture.

Another competitor: The 16-bit MT 68000 from Motorola had a 32-bit internal architecture, but it Did not manage to get out on time to become one of the possible options for the IBM PC.

In October 1978, six months after the announcement of Intel 8086, I moved to the MOS branch in TI and became a manager for the MP. By that time, everyone in the company, and many outside of it, knew that the 16-bit TI strategy for the MT was not working. Unsuccessful attempt of the department to develop a compatible 16-bit microcontroller TMS9940, which to my appearance was already 5 or 6 iterations, only increased this problem. I knew that I inherited a difficult situation. So why did I quit a good job as a manager of the Consumer Product Development Department? Location, location, location. The microprocessor enterprise was in Houston, and TI consumer products moved to Lubbock, Texas. Lubbock is a city in which the correct answer to the question "How do you like it here?" Will be "People here are wonderful." Country singer Mack Davies, who grew up here, once wrote a song with a chorus of "I thought that happiness is Lubbock, Texas in the rearview mirror."

Soon after I arrived in Houston, That I will have to give a presentation on the TMS9900 for a group from IBM working on a very secret project that required a 16-bit microprocessor. The group came from an unusual place for IBM: Boca Raton, Florida. I prepared for this very long, gave what I thought was a well-prepared presentation, and worked diligently with the consequences. But the team from IBM did not show much enthusiasm. Until 1981, we did not know what we had lost.

John Opel, president and then CEO of IBM, revolutionized the founding of the Boca Raton office, later known as Entry Systems Division. He realized that personal computers from Apple, Commodore, Radio Shack, TI and others could become a threat to IBM's dominance in the computer business. So he gave a group from Boca Raton, led by Philippe Estridge, a carte blanche to develop their product – a personal computer from IBM. They could use the help of third parties for anything, including the development of the OS and applications. This attitude made the system quite "open" by IBM standards, and accelerated the product's release to the market. However, Opel had one limitation: the product will be named IBM, so it can not undermine the company's reputation for quality and reliability. Therefore, a massive quality control organization inside IBM had to sign the output of this product before the sales began.

Instant success: IBM 5150 PC was released in August 1981. In the US, the price is $ 1565 Did not include a monitor, a printer and two disk drives.

There was no need to argue about the IBM team's choice of a 16-bit MP. Motorola 68K, as it was called later, was, without a doubt, a clear favorite. It had the largest logical address space, which was even more important than the minimum 16-bit internal architecture. It was easily extended to a fully 32-bit architecture. And, most importantly, 68K worked with the order of bytes "Big Endian", that is, from the oldest to the youngest. This is the order in which computers store bytes in memory. The 16-bit architecture came from an 8-bit architecture, and engineers had to decide which of the 8-bit bytes would be the first in the 16-bit world. Digital Equipment Corp. Chose the order from the youngest to the oldest ("Little Endian") for its Programmed Data Processor (PDP) and VAX architectures. Intel did the same. But IBM computers used a different order, "Big Endian". To "Big Endian" communicate with "Little Endian", the order of the bytes had to be reversed on the fly. At that time, such data conversion was not trivial. 68K from Motorola did not require conversion for use with IBM PC. So why do not we use computers based on 68K today?

The answer is related to who first entered the market. 8088 was not perfect, but, at least, it was ready, and 68K was not. The rigorous process of assessing the quality of IBM required that the manufacturer provide thousands of examples of the released product for each new spare part, so that IBM could conduct live tests on them. Hundreds of engineers have been involved in IBM quality control, but it takes time. In the first half of 1978, Intel had already produced examples of the 8088. By the end of 1978, 68K was not yet fully ready for release.

Unfortunately for Motorola, the Boca Raton group wanted to bring a new IBM PC to the market as possible Rather. So they had only two fully-ready 16-bit MPs to choose from. In the competition of two non-ideal chips, the Intel chip turned out to be less imperfect than the chip from TI.

TMS9900 did not quietly die without receiving recognition from the IBM PC. The managers still hoped to squeeze out the corporate strategy. Of course, the home computer from TI, the announcement of which is still expected, will have to use the TMS9900?

The computer development team reluctantly agreed to give it a chance. This group was the result of an unfortunate merger of two departments, one of which was developing a game console, and the other was a personal computer. The resulting hybrid was not adapted for either, for the other. But TI stubbornly bent his line. TI-99/4 entered the market in 1979, followed by TI-99 / 4A in 1981. The company eventually sold 2.8 million units, most of them at a loss, until the personal computer left in 1984 .

Meanwhile, the architecture of the Intel 8086 evolved and overcame the flaws. It still uses the order of the Little Endian bytes, but today it does not matter. And Motorola with its superior technology has lost the only major development competition in the last 50 years.

Since we are talking about the other competitors, I'll say a few words about the OS for IBM PC. The logical choice for a 16-bit OS was the expansion of the popular CP / M OS, developed by Gary Kildall in Digital Research based on the Zilog Z80. The Boca Raton group realized that there was an open standard for CP / M, so they ordered Digital Research to develop a version called CP / M-86. In the process, Microsoft came up with a proposal for MS-DOS, which has already been written a lot. Therefore, the PC world did not develop in the same direction that one could initially assume, both in terms of OS and in terms of the microprocessor.

What lessons can we learn from history? One – if you develop a product based on a rapidly changing new technology, then the main thing is to get first to the market, and it does not matter what limitations your initial product will have. Today in the Silicon Valley, this approach is called the creation of a "minimally viable product". If your product has new features that differentiate it from others, your customers will come up with innovative ways to use it.

The second lesson is if you manage a large corporation that wants to create a project in a small, isolated from other department, not burdened with luggage Or traditions, think about the restrictions imposed on it. It is likely that limiting the OS for IBM PC would bring the company more benefits in the long run than encumbering the product with complex quality control procedures. Nobody could predict how much personalities will affect our lives, but the real value is not in the hardware, but in the compatibility of the OS. If IBM, not Microsoft, controlled MS-DOS, Windows and others, the world of computers would be completely different.

Finally, for people standing on the sidelines and just watching the high-tech parade, I will say: not Miss your opportunities. In the case of TI, in 1979 we decided that the TMS9900 lost the general purpose MP race, and began to look ahead to the future, which will come after the general purpose MP. Our strategy focused on the MP narrow specialization and led to the development of the TMS320, a digital signal processor. Его анонс прозвучал на международной конференции по твердотельным схемам в феврале 1982 года, а вышел он в следующем году. Семейство 320 DSP и его производные начали приносить почти половину всей прибыли TI, подготовили новое поколение управленцев компании, и позволили TI участвовать в гонке встроенных процессорных систем на чипе. В 1990-х эта стратегия обратила вспять тенденцию к упадку среди лучших полупроводниковых компаний и принесла миллиарды долларов с продаж чипов для модемов, контроллеров дисководов и множества других продуктов.