Ten years of MCU history

Gary Boone, who was working in the MOS division of Texas Instruments, designed the first chip that could be called a microcontroller because he was bored with his job and his family was having troubles. He joined TI in 1969, just as calculator chips were becoming big business. In the 1960s, electronic calculators replaced the Marchant and Frieden electromechanical calculators that had dominated the market for decades. Semiconductors enabled the replacement of hundreds of complex metal and plastic components in electromechanical calculators, first by hundreds of transistors and diodes, and then by increasingly fewer integrated circuits. North America's Rockwell Microelectronics, Mostek, General Instruments and Texas Instruments are early players in the multi-chip calculator market.

Initially, dozens of integrated circuits were needed to replace hundreds of transistors and diodes. As more and more components are included in integrated circuits, fewer and fewer integrated circuits are needed to make a usable calculator. By 1968, IC-based calculator designs had largely replaced transistor-based designs. The end is obvious. Eventually, semiconductor manufacturers will reduce the electronic core of calculators to a single chip.

Japanese calculator suppliers Sharp, Canon and Busicom work with various U.S. semiconductor suppliers to develop custom chips for their calculators. Sharp partnered with Rockwell, Canon with TI, and Busicom with Mostek and Intel to develop different models of calculators. Busicom asked Mostek to develop a single-chip calculator and contracted Intel to develop custom chipsets for more complex programmable calculators. In late 1970, Mostek achieved this goal for the first time with the introduction of the MK6010, a custom chip that replaced 22 integrated circuits. Busicom integrated this chip into its small four-function desktop calculator, the Busicom Junior. The contract with Intel eventually led to the development of the Intel 4004 microprocessor. However, this story is about microcontrollers, which took a related but different evolutionary path.

TI's MOS division is deeply involved in calculator chipsets. Calculator companies including Canon, Olivetti and Olympia asked TI to develop 4, 5 and 6 chipsets for their calculators. Executing these custom chip projects fell to several TI engineers, including Gary Boone. The job required flying around the world to Japan, Italy and Germany. Boone spent a lot of time on the road, and his family resented his absence. Boone quickly grew tired of the stressful journey just to develop a new chipset that looked a lot like the one before it. In those days, many potential customers wanted calculator chips, but each customer wanted something a little different. That's the nature of the custom chip business. This is a customer-intensive industry.

Boone's frustration and family matters led him to Daniel Baudouin, TI's MOS marketing manager. Together, they compiled a matrix of customer needs from various calculator manufacturers. Then, they added a set of functional blocks that address those needs. Boone and Baudouin also noted what TI's MOS process technology can currently accomplish, and what it can do best. Their thinking quickly turned to architectures that made heavy use of memory (RAM and ROM). Because these structures were extremely efficient and easier to route on the IC, the memory promised to increase silicon utilization by 40 or 50 times.

Once Boone and Baudouin started thinking about memory, they started thinking about how much data and program storage a calculator chip would need. At that time, the TI team began discussing the prospect of a ROM-programmable single-chip calculator with potential customers. They got a lot of pushback. Customers accustomed to financing their own calculator chips balked at the idea of ​​calculator chips that were distinguished only by certain bits in on-chip ROM. There was also opposition within TI because ROM-based programmable parts were contrary to what the company was used to making.

Reading this far, you may notice that all the talk about calculator chips is inconsistent with the title. This series of articles is obviously about the history of microcontrollers. I assure you, we are not off track. The earliest microcontrollers were designed by Boone and TI engineer Michael Cochran. They included a processor, memory (RAM and ROM) and I/O all on a piece of silicon. They were calculator chips. They were the earliest microcontrollers. Take a look at the image below, excerpted from U.S. Patent 4074,351:

This figure shows the block diagram of TI's first microcontroller calculator, the TMS1802NC. It shows all the key components of a microcontroller. It has a CPU, which consists of a program counter (PC), an instruction register (IR), an instruction decoder (Control Decoders), and a 4-bit ALU. It has a RAM to store numerical data and a ROM to store programs that define the operation of the chip. Finally, at the bottom, you can see the dedicated I/O circuitry used to scan the matrix keyboard, drive the displayed digits, and drive the seven segments within each displayed digit. The I/O in this design may be specialized, but this diagram clearly depicts a microcontroller.

On September 17, 1971, TI released the TMS1802NC microcontroller calculator integrated circuit. Two months later, Intel released the 4004 microprocessor. Texas Instruments prices the device for less than $20. The ROM contains 320 11-bit instruction words (3520 bits), while the serially accessed 182-bit RAM contains three 13-bit BCD (binary coded decimal) numbers and 13 binary flag bits. In total, the chip requires about 5,000 transistors.

(Note: While researching this article, I found that more than one website confused TI's TMS1802NC 4-bit calculator chip with the 8-bit RCA CDP1802 COSMAC CMOS microprocessor released in 1974. The TI and RCA chips are not the same part , although the part numbers are similar).

A press release from TI from September 17, 1971 further confirmed the TMS1802NC’s status as a microcontroller:

"TI can easily implement any number of special operating features using single-level mask programming techniques using the same base or host design. The only limitations are the size of the program ROM, RAM storage and control, timing and output decoders. For example, by re- Programmed output decoder, the TMS1802 can be used to drive decimal displays such as Nixie type tubes."

One of the first calculators to use TI's TMS1802 calculator chip was the Sinclair Executive.

TI released the TMS0100 microcontroller calculator series on September 20, 1972, almost exactly one year after the release of the TMS1802C. The company renamed the TMS1802NC the first member of the TMS0100 family, TMS0102. Eventually, the family will have more than 15 different members, made on TI's 10-micron PMOS process technology. A year later, Mostek released the MK5020, an improved, pin-compatible copy of the TMS0102. Like all the semiconductor manufacturers listed at the beginning of this article, TI and Mostek would soon release microcontrollers developed in part using the knowledge gained from the creation of these early calculator chips.

Meanwhile, Boone has moved outside the calculator box. Calculator chip patent 4,074,351 describes other target applications including taxi meters, digital voltmeters, event counters, automotive odometers and measuring scales. Of course, microcontrollers have been used in all of these applications and much more.

Like many first-of-its-kind devices, Texas Instruments' TMS0100 calculator chip family is a narrowly defined class of microcontrollers primarily used to make calculators. However, the first chip in the TMS0100 family, originally called the TM1802NC and later renamed the TMS0102, contained everything a microcontroller needed: CPU, RAM, ROM, and I/O. Of course, this is a specialized microcontroller. Its I/O is application-specific and is designed to connect to a matrix keyboard and a seven-segment display. However, the TMS1802NC is a microcontroller.

Originally conceived by Gary Boone and Daniel Baudouin of Texas Instruments and then implemented by Boone and Michael Cochran, Texas Instruments introduced the TMS0100 series on September 20, 1972, and these chips quickly dominated the calculator market. TI suddenly had a whole new world to conquer. The company quickly realized that if the same programmable silicon could be designed to be versatile enough, it could serve multiple markets. TI applied its experience gained from its original programmable calculator chips to produce the first general-purpose microcontrollers, the TMS1000 series, released in 1974.

The TMS1000 microcontroller series has some similarities with the TMS0100 programmable calculator series, but there are also many differences. Both devices have 4-bit CPUs and Harvard architecture, which provides separate address spaces for RAM and ROM. Harvard architectures were common in early microcontroller designs because they simplified the design of the microcontroller's RAM, ROM, address decoders, and data buses. However, it seems to me that the Harvard architecture complicates the lives of programmers, who have to keep track of two different address spaces and often have to devise ways of moving data from ROM to RAM. (Fortunately, there is no point in moving data from RAM to ROM, the data will not complete the process. You cannot successfully write to a masked ROM.)

Unlike the TMS0102's 3250-bit ROM, organized into 320 11-bit words, the first TMS1000 microcontroller had a 1-kilobyte ROM organized into 1024 8-bit words. Therefore, the TMS0100 and TMS1000 series started with incompatible 11-bit and 8-bit instruction sets respectively. Similarly, the TMS0102's 182-bit serial RAM contains three 13-bit numbers in BCD (Binary Coded Decimal) format and 13 binary flags, while the TMS1000 microcontroller has a 256-bit RAM organized into 64 4-bit words.

According to the TMS1000 documentation, the 64 4-bit words stored in RAM are "conveniently grouped into four 16-bit files, addressed by a 2-bit register." In my experience, when writing similar microcontroller architectures (discussed later in this series), there is no convenience in further dividing a small RAM address space into 16-word chunks, unless you are writing 16 entries circular buffer. Like many design compromises, the microcontroller design team crammed the entire CPU along with RAM, ROM, and I/O circuitry onto an early semiconductor die, and I bet this particular design choice made the hardware design of the TMS1000 microcontroller Simpler and smaller, often sacrificing programmer convenience for hardware design convenience.

Unlike the dedicated I/O design of the TMS0100 calculator chip family, the TMS1000 microcontrollers have general-purpose I/O pins, at least in name only. The four input pins (K1, K2, K4 and K8) can be read as a group with one instruction. Output pins are more complex. The original TMS1000 microcontroller had 11 "R" outputs (R0 to R10) and 8 "O" outputs (O0 to O7). The "R" outputs are set and cleared respectively. The "O" output is controlled by a mask-programmed PLA and driven by a 5-bit latch. The four bits in the latch can be set with an instruction that moves data directly from the TMS1000's accumulator to the latch. The fifth output bit comes from the ALU's status latch. The use of PLA to extend the 5-bit output latch to 8 output pins is reminiscent of the TMS1000's heritage as a descendant of the calculator chip, which was designed to drive a 7-segment display.

Among Adam Osborne's many achievements, he documented early microprocessors and microcontrollers in his 1978 book "Introduction to Microcomputers." In his description of the TMS1000, Osborne seemed more concerned about the limitations of the microcontroller:

"The fact that the TMS1000 series microcomputers are single-chip devices has some minor, unobvious implications. Most importantly, there is no such thing as a support device with 1024 or 2048 bytes of ROM [The TMS1200 microcontrollers have 2Kbytes of ROM ] represents the exact amount of program memory that will appear; no more, no less. Similarly, 64 or 128 bytes of RAM (a byte is a 4-bit word) cannot be extended - no direct memory access logic exists. And its existence makes little sense anyway; since the total amount of RAM and ROM available is so small, there's simply no chance of transferring a block of data long enough to warrant bypassing the CPU.

Likewise, in the TMS1000 microcomputer, the role of interrupts is negligible. Given the small amount of in-program memory available and the low cost of the program packages, it was difficult to justify the complexity of interrupt logic just to have the microcomputer perform multiple tasks. "

To me, Osborne's words indicate that many people (probably including Osborne) did not clearly understand the difference between microcontrollers and microprocessors when he published that book in 1978, four years after TI first announced the TMS1000 series. difference between. However, many people did understand the difference because by 1979, TI was reportedly producing tens of millions of TMS1000 parts per year, and they were selling TMS1000s in bulk for $2 or $3. The TMS1000's low unit cost is possible in part because TI packages the device in an inexpensive 28-pin plastic DIP.

TI is eating its own dog food by using members of the TMS1000 microcontroller family in some of its own consumer products, including the legendary TI Speak & Spell game and the SR-16 "Electronic Slide Rule" calculator.

Inventor, game designer, and "Father of Home Electronic Game Consoles" Ralph Bell realized he could make affordable video games with microcontrollers and integrated the TMS1000 into one of the most successful handheld electronic games - — Milton Bradley's Simon, released in 1978. Today, everyone plays handheld games on their phones, but back then, those games required specialized hardware.

Parker Brothers released Merlin, a handheld video game based on the TMS1000 microcontroller, in 1978, and a year later the Milton Bradley Company used the TMS1000 microcontroller as the programmable brain of its Big Trak, a futuristic 6 The wheeled tank-like vehicle can be pre-programmed to follow a specific path via a membrane keypad embedded in the back of the toy. Big Trak can execute a sequence of 16 commands typed into the keyboard, which appears to be similar to the Logo programming language developed in 1967 by Wally feurzeeig, Seymour Papert, and Cynthia Solomon at a research firm called Bolt, Beranek, and Newman (BBN) in Cambridge, Massachusetts. The turtle graphics are closely related.

In 1977, Mattel launched a very successful electronic football game. The game was based on a Rockwell calculator chip, but companies around the world cloned the game, and a Hong Kong game maker called Conic appears to have used a TMS1000 microcontroller in its clones instead of the calculator chip. The open source game emulator MAME (multi arcade emulator) can still run the TMS1000 ROM code of "Conic's Football" in the simulation.

Author Stan Augarten noted in his book "The State of the Art" that the TMS1000 was used in calculators, toys, games, appliances, burglar alarms, copiers, and jukeboxes. Augarten concluded his description of the TMS1000 by writing: "Like any integrated circuit, the TMS1000 helps make the power of modern electronics available to everyone."

I suspect there are countless undocumented TMS1000 family applications out there. It's quite a success story and legacy for a second early microcontroller family, and a testament to the true universality of the basic single-chip microcontroller concept. After TI launched the TMS1000 in 1974, new microcontrollers launched by other semiconductor manufacturers appeared at a rapid rate.

Now that we have entered the 21st century, most people rarely think of Rockwell Microelectronics' connection to microprocessors and microcontrollers. Parent company Rockwell North America (renamed Rockwell International Corporation in 1973) is a major military/aerospace contractor. Rockwell built the Apollo spacecraft, the B1 bomber and the U.S. space shuttle.

For a long time, most of the United States' space booster rockets and intercontinental ballistic missiles have used Rockwell's Rocketdyne engines. In 1972, Rockwell launched the world's third commercially successful microprocessor, the 4-bit PPS-4. In 1976, Rockwell released a single-chip microcontroller based on the PPS-4 architecture. It's called PPS-4/1.

As was the trend among many large conglomerates in the 1960s, Rockwell started its own semiconductor manufacturing operations in 1967 in its Autonomous Division. Autometics develops various military/aerospace avionics systems, including inertial navigation and guidance systems for U.S. submarines and intercontinental ballistic missiles, creating demand for advanced semiconductors. North American Rockwell Microelectronics Corporation (NRMEC) developed early MOS/LSI process technology for its military and aerospace projects.

When Japan's Sharp turned to a semiconductor supplier to manufacture a calculator chipset of its own design, NRMEC's ​​MOS/LSI capabilities met Sharp's needs. The result of the collaboration is a quad chipset that Sharp incorporates into its QT-8D calculator. Sharp released this calculator in August 1968. In fact, one might say that the Rockwell chipset in Sharp's QT-8D calculator ushered in the MOS/LSI era. In 1970, Rockwell began publishing a MOS/LSI chip catalog.

As Texas Instruments discovered with the TMS1000 microcontroller, it was only a brief transition from electronic calculator architecture to 4-bit microprocessors, or in the case of Texas Instruments, to microcontrollers. jump. Rockwell announced the 4-bit PPS-4 microprocessor family in August 1972. It was the world's third commercially successful microprocessor, following Intel's release of the 4-bit 4004 and 8-bit 8008 microprocessors. Rockwell's "PPS" name means "Parallel Processing System."

Two things set Rockwell's PPS-4 microprocessor apart from its competitors. The first is Rockwell's unique QUIP (Quad Inline Package) device package. Rockwell's QUIP chips are easily identifiable by their staggered leads. Because of their appearance, these chips are often called "spiders." In an era when minimum board traces and space are on the order of 10 mils, the QUIP lead configuration makes it easier to design printed circuit boards for these devices.

第二个显著特点是罗克韦尔为PPS-4微处理器开发的配套芯片。到1975年，芯片组家族包括CPU、罗克韦尔独特的四相时钟所需的时钟生成器/驱动程序、256×4-bit RAM、1和2千字节ROM、RAM/ROM组合芯片、键盘和显示控制器、打印机控制器、通用I/O芯片和1200 bps模拟调制解调器。(1200 bps模拟调制解调器开启了一长串调制解调器芯片，导致NRMEC成为Conexant Systems，最终被Synaptics收购。)

The large number of chips in the Rockwell PPS-4 microprocessor family enables the processor to be used in a wide range of end products, including cash registers, fax machines, home appliances, pinball machines, toys and calculators. However, the market for multi-chip, 4-bit microprocessor families was short-lived. Semiconductor technology is developing rapidly, and device density is increasing at an alarming rate. In October 1975, Rockwell integrated the clock generator into the microprocessor and merged the RAM, ROM and I/O peripherals into a 2-chip set called PPS-4/2, but due to the semiconductor process With further advancements in technology, 2 chipsets are also available in a short period of time. In early 1976, Rockwell released the PPS-4/1, a true monolithic microcontroller based on the original PPS-4 microprocessor architecture.

There is little history of Rockwell PPS-4 applications on the Internet, except for information about the PPS4-based calculator and one rather unusual application: the Gottlieb pinball machine. Gottlieb contracted with Rockwell to develop a System 1 pinball controller board based on the PPS-4/2 microprocessor. Gottlieb used System 1 boards in pinball machines released from 1977 to 1980. The first pinball machine to use System 1 boards was called "Cleopatra." Other microprocessor-based pinball games followed, such as Sinbad, Dragon, Charlie's Angels, Incredible Hulk, Buck Rogers, and Totem. There are 16 pinball games in this series.

It's the kind of history that's easily forgotten by time, but the PPS-4's history as a pinball controller has not been, for two reasons. First, collectors are rewarded with a Gottlieb pinball machine based on the System 1 board. Second, the metal gate, MOS/LSI PPS-4 rom/peripheral chips on these boards are failing, now that they are approaching half a century old. Typically, pinball collectors are stuck on scrapped machines after these parts fail because they haven't been produced in decades and semiconductor supplier NRMEC is long gone.

French consultancy AA55 has developed a solution to this problem. The company developed the FPGA-based Rockwell PPS-4 peripheral chip. AA55 Consulting's FPGA code targets appear to be produced by AMD/Xilinx, as the project files are formatted for Xilinx's ISE development software. AA55 Consulting has not yet reverse-engineered the PPS-4/2 processor, but plans to do so in the future. Maybe next year.

Since Rockwell components are now almost pure unobtanium, NI-Wumpf in Honeoye Falls, New York, has developed a functional replacement for the original Gottlieb System 1 board that does not use any Rockwell semiconductors. The original NI-Wumpf board appears to be based on the Zilog Z80 microprocessor. The latest version uses the STMicroelectronics STM32F103 microcontroller, which contains a 72 MHz Arm Cortex-M3 processor. For comparison, the PPS-4/2 microprocessor operates at 199kHz. A French company called Flippp! has taken a similar approach, developing board-level replacements for System 1 boards called the PI-1 and PI-1×4, designed and programmed by Pascal Janin. The board also appears to be based on a more modern microcontroller.

It is interesting to read some of the reviews from users of the Gottlieb pinball machine based on the System 1 board. Most people think of Rockwell as strictly a defense contractor. A website called pinwebsite has a forum dedicated to Gottlieb, and one of the threads is "Why is Gottlieb's System 1 so bad?" Here are some quotes from the thread:

"At the time, Rockwell seemed like a good choice. After all, they designed computer equipment for NASA and the Department of Defense, so what could go wrong? Apparently, there were a lot. Rockwell decided to do the grounding thing There were questionable decisions, using custom-designed components and other weird stuff that really screwed Gottlieb up."

"Why would Rockwell spend extra time and effort designing custom Spider chips when they are still inferior to the off-the-shelf 68xx chips used by William and Bally?" It seems unlikely that going with custom hardware would save them any cost. "

Gottlieb was unaware and made some poor choices during the outsourcing process. As for Rockwell, you might wonder what was the greater miracle: the success of the pinball machine, or the success of their NASA product?"

I think Rockwell used parts that were "off the shelf" for them with the spider. Part of the problem may be that they never designed the hardware for the environment in which pinball was played. Expect the switch to bug out when turned on instead of slamming off? Isn't the hardware "off the shelf" to anyone outside the defense industry?

"I don't want to get too political, it's just my opinion, if I want to hire a company to design some electronics for me, even if Rockwell has a good name, I don't want to hire a company that's used to doing government projects. In For the most part, they were bloated, expensive, and over-engineered. I'm not sure that was the case in the late 70s, but just because a space device worked well, it didn't mean it wasn't overpriced and over-engineered. "

"Rockwell designed the system as a 4-bit system - which was already obsolete before it was released."

There is no doubt that people who collect pinball machines as a hobby today do not know that Rockwell was a commercial chip supplier in the 1970s, do not know that the PPS-4 has a long history, and do not know why the 4-bit microprocessor was developed. processors and microcontrollers, no idea why Rockwell developed the "Spider" QUIP, or half a century of poorly maintained systems tending to fail. However, some collectors are well-informed. For example:

"Electronically - the System 1 used mid-70s technology. All electronics were off-the-shelf components, no custom components. In fact, Rockwell used their own parts - who could blame What about them? I think it has only two drawbacks - poor grounding techniques and edge connectors.

"As for the unpopularity of their CPUs - they were very popular in point-of-sale terminals for quite some time. But when MOS Technology introduced the affordable 6502 series of processors, their 4-bit processors fell by the wayside."

Rockwell Microelectronics and the PPS-4 product line have also faded from the collective memory of those actively involved in the electronics industry, and the history of the PPS-4 is difficult to find online. You almost have to turn to old books. Fortunately, I have some of these books on my study shelf.

For example, the 1981 edition of the Osborne 4 & 8-Bit Microprocessor Handbook lists 10 members of the Rockwell PPS-4/1 microcontroller family. Family members have 640 to 2048 bytes of ROM and 48 to 128 4-bit RAMs. All but one family member has three integrated serial I/O ports, which are essentially nothing more than serial 4-bit shift registers. The MM76C is a family member with a high-speed upstream/downstream timer/counter subsystem that can operate as a 16-bit counter or two 8-bit counters. The counter can also handle the quadrature encoding input used by optical encoders. The timer/counter subsystem opens up additional industrial application areas for Rockwell PPS-4/1 microcontrollers, including motor control, frequency counting, analog-to-digital conversion, and frequency synthesis.

If you're not familiar with NRMEC, Rockwell Microelectronics, or Rockwell Semiconductor, that's probably because the company was spun off in 1999 as Conexant Systems, a global push to increase the value of internally held semiconductor companies. part of the effort toward the stock market. Conexant split its wafer fab (formerly Rockwell's wafer fab) into Jazz Semiconductor in 2002 and has since moved towards fabless. Tower Semiconductor acquired Jazz Semiconductor in 2008 and became TowerJazz. The company revived the Tower Semiconductor name in 2020, and now Intel is acquiring the company. Meanwhile, Rockwell's early MOS/LSI process was gone forever and mostly forgotten.

Intel announced the 4004, the first commercially successful microprocessor, in late 1971. By 1974 Intel had introduced four microprocessors: the 4-bit "low-end" 4004 and the upgraded 4040, and the 8-bit 8008 and 8080. Intel's 4-bit 4004 and 4040 microprocessors are primarily used in embedded control applications where I/O functionality and performance as well as lower parts cost outweigh the superior processing performance of 8-bit devices. However, Intel no longer has a monopoly on the microprocessor market. Several other semiconductor vendors introduced competing microprocessors in 1974, notably the 4-bit Rockwell PPS4, the 8-bit Motorola 6800, the multichip Fairchild F8, and National Semiconductor's 16-bit multichip ( multichip) IMP-16. Toshiba designed, manufactured and delivered the 12-bit TLCS-12 microprocessor specifically for Ford's engine controller. At the same time, there are many MCUs on the way.

更糟糕的是，德州仪器 (TI) 于 1974 年推出的 4 位 TMS1000 微控制器系列将 4 位 CPU、RAM、ROM 和 I/O 电路放在一个芯片上，从而简化了系统设计并显著降低了基于处理器的控制成本。TI 的 TMS1000 系列的推出当然引起了一些潜在客户（以及一些半导体制造商）的注意。处于围攻之下的英特尔开始将嵌入式系统设计输给嵌入式市场中的微处理器和微控制器竞争对手。

This has to stop. Intel needs to respond, and the company knows it.

When Henry Blume Jr. came to Intel in October 1974, there was an agreement within the company to develop microcontrollers. Intel already makes microprocessors, RAM chips, ROM and UV-erasable EPROM and has the process technology to make these parts in-house. But the main decision they haven't made yet is whether the microcontroller will have a 4-bit CPU (like the TMS1000) or an 8-bit CPU. Based on discussions by the Computer History Museum (CHM) in 2008 with many of the people responsible for developing the 8048, Ed Gelbach, Intel's senior vice president of corporate marketing, decided that Intel's microcontroller would have an 8-bit CPU because 4-bit CPUs did not Attractive enough.

Although 8048 is the common name traditionally used for Intel's first microcontrollers - the actual series name was MCS-48 - the oral history mentioned above also makes it clear that the 8748 EPROM version came first. This is because the 8748 is immediately available for software development and early prototyping, whereas customers would need a year or two to develop their software before ordering a ROM-based 8048 device. The 8748 will also be a more difficult device to manufacture because Intel's EPROM process technology is very different from the company's other process technologies, so it would make sense to have that device ready first and then design a ROM-based 8048 in the coming year. Finally , Intel can and is willing to charge more for the reusable 8748, which means the company can generate more revenue faster.

David Stamm joined Intel in January 1974 and initially worked on fixing bugs in the 4004 and 4040 microprocessors then in production. He then designed the Intel 4308, a chip that supported the Intel 4040 and combined 1 KB of ROM and some I/O ports. The 4308 combined the functionality of four Intel 4001 chips that combined a 256-bit ROM and an I/O port, so the 4308 contained about half a microcontroller, lacking the CPU and RAM. After designing the 4308, Stamm was assigned to the 8048 project. He said in 8048's oral history that he was single at the time, so he "spent more than two years living and breathing 8048, day and night."

Stamm's first responsibility was developing the microcontroller's instruction set. This will be completely new and fit into an 8-bit microcontroller, not the microprocessor Intel already makes. In the oral history, Stamm recalled:

“There are three of us working full-time on this project: myself, who is responsible for the instruction set and logic design and overall chip schedule; and David Buddy, who is responsible for everything related to integrating the EPROM technology, all EPROM programming logic and sensing logic for the EPROM components, and most of the circuit design Complex tasks; and then there’s Dwayne Hook, who is responsible for all the layout of the chip…

"The first phase is really instruction set design. So here I am, I'm a year out of college with my bachelor's degree, and I'm thinking what business do I have in developing the next generation of instruction sets for these components? Really, no one is paying close attention looked over my shoulder, but I said, okay, fine, I'll do it. Luckily, in college I learned assembly language programming in a language called COMPASS, which is part of the CDC, It no longer exists."

"But I learned a lot about assembly language programming there, and then I looked at the 4040, and I looked at all the other instruction sets to try to figure out which instruction set might make sense. The challenge is, you have to base the cost of the chip on the complexity of Consider the instruction set based on the nature of it.”

“因此，例如，减法和比较将是非常有价值的指令，但它们会显著增加额外的芯片面积。所以，我决定我们真的负担不起这些，特别是因为我们要进入 8 位设计。我一直是 8 位的大力支持者。反对意见之一是成本和芯片尺寸。因此，在这些设计步骤中，我尽我所能放弃我认为会增加额外芯片尺寸的特性和功能。现在回想起来，我想我可能有点过火了——尽管当时很难得出这样的结论。”

At one point in the oral history, Blume interjected:

"I also want to point out that two of Dave's famous commands - or his favorite commands were SEX and SIN, which stood for Set External Mode and Set Internal Mode. And then when the people in the system took over, they deleted SEX and SIN … They renamed them.”

The design limitations of the 8048 were to ensure that the chip could be manufactured using available process technologies, but were not limited to the instruction set. Memory space limitations are another important limitation. The original 8048 had a 4 KB program memory address space limit, but it was actually split into two 2 KB memory areas. The 8048's program counter is ostensibly 12 bits, but the most significant bit comes from a separate register that can be used for bank switching. Originally, the 8048 contained a 1 KB of program memory, so the decision was not to pinch initially. Eventually, it will. “…there are many, many limitations, primarily to reduce chip size,” Stamm explained.

Intel didn't have simulation software in its early days, so the design team built breadboards instead. Stamm recalled:

"So we built this big breadboard, which was a complete design project in itself, using the same type of combinational logic that we were hoping to eliminate by developing the 8048... We were using TTL and DTL devices, and I remember the breadboard was very Big. It was about five feet tall and two or three feet wide, but the back was completely covered in wires."

This was the state of the art in chip design at the time.

Before the new microcontroller could be put into production, the design team had to meet with Intel executives—Andy Grove, Gordon Moore, and Lex Vadasz—and convince them that the 8748 was ready for the market. With his demo ready, Stamm decided to write a blackjack program. Stamm enjoyed gambling and frequented Lake Tahoe casinos. The 8748 is programmed to drive a dumb terminal, presumably through an RS-232 voltage converter. Stamm recalls discovering the limitations of his microcontroller's instruction set while writing a blackjack program. He also filled up the 8748's program memory and there was not enough room to add a double down feature to the blackjack game. The first applications written for the 8048 immediately revealed its major limitations.

Intel announced the MCS-48 microcontroller family in late 1976, which included the 8048 and 8748, with the goal of shipping 1,000 revenue units, or all 8748 devices, in the first quarter of 1977. Blume recalled that the actual number of revenue units shipped was 770, which everyone considered a success, no doubt the result of heavy pre-sales. The Intel 8748 was a huge hit. Intel product manager Howard Raphael recalls early customers as Gilbarco (gasoline pumps), Tektronix and Chrysler.

Magnavox's Odyssey 2 video game console is based on the 8048. The 8048 is widely used to drive a variety of analog music synthesizer keyboards, including the Korg Trident series, Korg Poly-61, Roland Jupiter-4, and Roland ProMars.

The Sinclair QL personal computer uses an Intel 8049 (8048 with 2 KB ROM) to manage its keyboard, joystick port, RS-232 input, and audio output. Nintendo used a ROM-less 8035 microcontroller (most likely an 8748 with a bad EPROM) in its original Donkey Kong arcade game to generate the game's music. My friend and colleague Wally Wahlen used the Intel 8048 as the controller in his design for the Hewlett-Packard 9876 thermal page printer introduced in 1979.

Eventually, IBM PCs would use the Intel 8048 as their keyboard controller. However, the IBM PC was not the first computer to use the 8048 in this way. This milestone belongs to the Tandy TRS-80 Model II, which uses a reduced-cost version of the 28-pin 8048 called the 8021 to manage its detachable keyboard and scan keys.

The Intel 8048 series was a huge success for Intel, but its design limitations became apparent almost immediately. By 1977, program address space limitations were beginning to clearly limit customers, and by the fourth quarter of that year, just a year after launch, Intel began defining the 8048's successor, the 8051. It will become an even bigger success.

Motorola's semiconductor components group - Motorola Semiconductor - was late to the game with microprocessors for more than one reason. The company never developed successful PMOS process technology, so it sent custom LSI chip designs (such as calculator chips for customers) to other semiconductor manufacturers such as Mostek and AMS (Advanced Memory Systems) for manufacturing. Although it is a leader in bipolar IC manufacturing with a strong line of RTL, DTL, TTL and ECL logic chips, the company lacks PMOS LSI process technology, so Motorola cannot solve the design and manufacturing of large chips such as microprocessors. But they ended up developing the 8-bit microprocessor, the Motorola 6800, a well-thought-out architecture that became a huge success as a microprocessor and then for decades more as the foundational architecture for several microcontroller families.

C. Lester (Les) Hogan left Motorola Semiconductor in 1968 to become president of Fairchild Semiconductor. Seven Motorola Semiconductor executives (all but one) resigned to follow Hogan, who quickly found them new positions at Fairchild. Initially, Hogan was not interested in running Fairchild, so he turned down the position when it was first offered. Motorola's semiconductor business has surpassed Fairchild Semiconductor. In addition, Fairchild is funneling talent to other semiconductor companies and startups in Silicon Valley. It's an ugly situation, and Hogan knows it. So Fairchild sent its founder, Robert Noyce, to Phoenix, Arizona, to recruit Hogan. Hogan recalled that he was an excellent salesman.

Hogan accepted the invitation and negotiated a large (for the time) compensation package for himself. He and seven of his top executives left Motorola and headed to Silicon Valley to try to stop Fairchild's bleeding. Ironically, Noyce left Fairchild Semiconductor shortly after Hogan joined to found Intel with Gordon Moore. Noyce was reportedly upset that he had not been offered the top job at Fairchild Semiconductor, the company he co-founded in 1957, and he had let Hogan know he was leaving before Hogan accepted the new position. Hogan and his team successfully righted the ship of Fairchild Semiconductor, but the departure of almost all senior executives left Motorola's semiconductor group adrift and its strategic plan in disarray.

This management change in 1968 undoubtedly hindered key development projects at Motorola Semiconductor, such as the development of efficient PMOS process technology. The lack of such a process meant they couldn't have any ambitious LSI projects on their books, except for a few custom LSI projects brought in by the sales force and marketing department.

Three years later, in 1971, Tom Bennett joined Motorola Semiconductor to help the company enter the calculator chip business - the hottest LSI chip business at the time. It was this business that led Texas Instruments to develop the TMS1802NC programmable calculator chip, the first that could be called a single-chip microcontroller. He had a considerable background in computer design and in 1969 saw early design ideas for the Intel 4004 microprocessor. When Bennett joined Motorola, the company had developed a scrappy, bottom-up approach to project development that was sufficient to allow the development of NMOS LSI process technology and the 6800, Motorola's first microprocessor.

Jeff LaVell joined Motorola in 1966 after working on the company's C8500 computer at Collins Radio, where he learned about the development of computer architecture. Collins' C8500 prototype uses Motorola's ECL logic chip, and the production version uses Sylvania's SUHL TTL chip, with Motorola as the second supplier. So when LaVell and his wife decided to move to Phoenix, he was very familiar with Motorola. He joined Motorola Corporation's computer industry marketing organization as an applications engineer.

Part of LaVell's job involves working with major computer companies such as CDC (Control Data Corporation), DEC (Digital Equipment Corporation) and Cray, and understanding where semiconductor companies like Motorola can help. A huge opportunity emerged: computer peripherals. The device requires highly integrated functionality to minimize cost, chip count, design time and power consumption. This market is well suited for microprocessors with a complete family of supporting chips.

Bill Lattin received his master's degree in 1969 from the University of California, Berkeley, where he audited Professor Andy Grove's semiconductor physics course at Intel. He had planned to move to Phoenix, complete his PhD at Arizona State University, and take a position at Motorola Semiconductor, eventually joining the company's MOS process development group after rotating through several other departments. He became the group's design manager, responsible for the development of the company's CAD tools and new MOS chip designs. He also took on Motorola's issues in existing MOS process technology and new MOS process development.

Bennett, LaVell and Lattin worked together to define a microprocessor and a series of supporting devices that would be fabricated using as-yet-undeveloped NMOS process technology. The approximately 15 chips they defined include:

• MC6800 microprocessor

• MC6810 128-byte (1-Kbit) RAM

• MC6830 2-KB ROM

• MC6820 PIA (Parallel Interface Adapter)

• MC6850 ACIA (Asynchronous Communication Interface Adapter)

• MC6852 SSDA (Synchronous Serial Data Adapter)

• MC6860 600-bps digital modem

The Motorola team visited a long list of companies, including several HP divisions, two TRW divisions, NCR (National Cash Register), and the CDC, to sell the idea of ​​a new Motorola microprocessor chipset. They started to get attention. CDC may order 200,000 parts to make the project a reality. Logic design started in 1972. The layout of the first five chips (microprocessor, RAM, ROM, PIA, and ACIA) continued until 1973. The first batch of mostly working but with some errors arrived in February 1974. Customers had the chip working in their systems by June 1974.

Transcripted from the Computer History Museum's Motorola 6800 oral history panel, one of Bennett's early customer visits really caught my attention:

"In June, I went to HP in Loveland and saw a system I think on a Friday afternoon. It was amazing how much they narrowed that thing down - because they [showed] what they had before and after. They were getting ready to go to a park there that night because there was a Coors truck loaded with their beer. Then we got on the plane and came back and they went on so fast."

Exactly a year later, I joined the HP division's lab, the Loveland Calculator Products Group. The prototype system described by Bennett in his oral history was apparently the prototype for the CJ project that became the HP 9815A programmable desktop calculator and instrument controller introduced in 1975.

The MC6800 microprocessor chipset was an immediate success. It is not a microcontroller, but multiple chips combined to provide the functionality of a microcontroller. All that is required is integrating the five initial devices in the chipset onto a single piece of silicon. Motorola announced the microcontroller MC6801 in 1977. At the request of major customer General Motors (GM), the MC6801 featured a slightly upgraded 6800 CPU that included a multiply instruction. GM originally used the MC6801 microcontroller for the TripMaster digital trip meter in the 1978 Cadillac Seville. This is a very expensive – $920 option.

TripMaster brings Motorola into the ranks of General Motors, which has developed a microcontroller-based carburetor closed-loop control system to help its vehicles meet new government standards for fuel consumption and emissions. By the second half of 1980, General Motors was manufacturing 25,000 Motorola microcontrollers per day in its cars, which helped Motorola Semiconductor survive the 1980/1981 recession.

The Motorola 6801 microcontroller spawned a series of architecturally similar microcontroller families, including the MC6805, the CMOS version of the MC6805, the MC146805, MC68HC05, MC68HC08, MC68HC11, and MC68HC12. As process technology improved, Motorola followed Moore's Law and introduced microcontrollers with more RAM, more ROM, more on-chip peripherals, and more I/O functions. These microcontrollers are used in pagers and mobile phones made by Motorola, the parent company of Motorola Semiconductor. Motorola ships a lot of microcontrollers. Eventually, Motorola spun off its semiconductor division into Freescale, which merged with NXP in 2015. The NXP website still lists several microcontrollers derived from the original MC6800 microprocessor architecture.

The MC6800 microprocessor also has a microcontroller legacy worth mentioning. Motorola needed a second source to increase trust in parts for some customers, and they chose AMI (American Microsystem Inc) as their initial second source. In Japan, Hitachi also became a second source of MC6800, helping the company enter the microprocessor business. Eventually, the company developed a low-power CMOS version of the Motorola MC6801 microcontroller, called the HD6301, which spawned a number of new microcontroller families now sold under the Renesas brand.

When Federico Faggin came to Intel in 1970, he immediately found himself in chaos. He left Fairchild Semiconductor and accepted a position at Intel, where he was introduced to Busicom's custom chipset project, which eventually became the 4004, the first commercially successful microprocessor. Faggin developed silicon-gate MOS process technology at Fairchild, knew it was far superior to the metal-gate technology everyone was using at the time, and by observation had left Fairchild to follow Bob Noyce and Gordon Moore to their new semiconductor startup Intel. Faggin also wanted to join because Fairchild had ignored his new process technology, so he called his old Fairchild boss Les Vadasz (who had gone to Intel by then) and asked for a position, which Vadasz quickly agreed to.

On April 3, 1970, the day Faggin joined Intel, Stan Mazor handed him all the documentation for the Busicom project. Ted Hoff and Mazor have developed the high-level architecture and instruction set for the 4004 microprocessor, leaving only logic design, circuit design, and silicon-gate process development to be completed. In other words, much of the project remains unfinished. In addition, Mazor informed Faggin that Busicom engineer Masatoshi Shima, who was overseeing the project on the customer side, would arrive from Japan the next day to obtain a progress report.

When Shima arrived the next day, he was eager to check the progress and asked to be taken to Intel immediately to see the documentation. He wanted to see the progress of chip logic design. But after learning there was no progress, Shima became furious. Faggin calmed Shima down, and Shima called Busicom in Tokyo. The entire project is in jeopardy as there has been no progress on the project over the past few months.

In this case, Faggin took over the development of Intel's 4004, 4040, 8008, and 8080 microprocessors. By 1974, he was probably the world's foremost expert on microprocessor development practices, from architecture and instruction sets to process technology, and he was preparing to leave Intel. Bob Noyce and Gordon Moore handed over more and more day-to-day control of the company to Andy Grove, whose confrontational management style frustrated Faggin. In addition, Intel was benefiting from the huge DRAM wave at the time, and Andy Grove was not particularly interested in microprocessors at the time, which he thought would be a distraction.

Then, Faggin discovered that Intel had applied for a buried contact patent for Faggin's silicon-gate process, which Faggin invented, but Fairchild had not applied for a patent. To make matters worse, Intel completed the work without Faggin's knowledge and placed the patent in the name of Faggin's boss, Vadasz. The discovery of this behavior seemed to be the final straw for Faggin, who had long felt that he was underappreciated by Intel's management and felt that he was grossly undercompensated for his significant contributions to the company's microprocessor projects.

By the end of the summer of 1974, Faggin decided to leave Intel and start his own company. He invited one of the managers who reported to him at Intel, Ralph Ungerman, out for a drink. Ungerman took over Intel's microprocessor business from Faggin in early 1974. Faggin asked, "Ralph, how about you and I start a company - a microprocessor company?" Ungerman replied, "Okay!" and that was the beginning of Zilog.

Faggin's last day at Intel was October 31, 1974 - Halloween. He wanted to make a clean break with Intel, so he didn't develop a business plan or arrange funding. He had just come up with the idea of ​​starting a microprocessor company, and the first product he wanted to develop was a microcontroller. He called it 2001. Ungerman started a consulting firm called Ungerman Associates on the side while working at Intel. When they became partners, Faggin bought half of that company. Ungerman's consulting income initially funded the new company, which was eventually renamed Zilog.

Faggin spent November and December developing the architecture and instruction set for the 2001 microcontroller. During that time, Electronic News, the leading industry newspaper of the day, ran an article about Faggin's new microprocessor venture. The article caught the attention of a partner at Exxon Enterprises, the oil company's investment arm. Fagin received a call from Exxon and asked to meet in a few days. Faggin agreed, asked his wife to write a brief business plan, and then they met. Faggin presented his plans and initial thoughts on the 2001 microcontroller. The Exxon corporate partner said he was interested enough to continue the conversation.

Then, in December, Faggin realized that the microcontroller business wasn't big enough to serve as the basis for his new microprocessor company. One Saturday in December 1974, he suddenly thought of the right product. "Super-80" he exclaimed. His vision was a Super-80 microprocessor based on the Intel 8080 architecture, but with many improvements such as twice the registers, more instructions, more addressing modes, more bit-level instructions, and better interrupts structure, and peripheral chips designed from the outset as a series. In addition, the Super-80 will be fabricated using an NMOS process that requires only 5 volts of power, so it will be much easier for system designers to use. The Super-80 microprocessor will have an immediate and larger target market than the 2001 microcontroller.

Faggin called Exxon's investment partners to inform him of the change in strategic direction. Exxon adopted the idea, and by April of the following year, Faggin received a verbal commitment of $500,000 from Exxon. The 2001 microcontroller idea was shelved and work on the Super-80 began. The Super-80 released by Zilog in July 1976 was the later extremely successful Z80 microprocessor.

Even before Zilog announced the Z80 microprocessor, Faggin began thinking about the company's next-generation microprocessor once the company had the parts available. He wanted to increase and decrease complexity at the same time. To increase scale, he wanted a 16-bit processor, which became known as the Z8000. In early 1976, Faggin hired Dr. Bernard Peuto, a computer architect who had developed IBM-compatible mainframe architecture at Amdahl, to serve as the architect of the Z8000 microprocessor. In an effort to downsize, Faggin took plans for the 2001 microcontroller off the shelf where it had sat for more than a year, dusted it off, and renamed it the Z8. Faggin and a committee of Zilog engineers, including Peuto, improved the original architecture of the 2001 microcontroller.

At Intel, Faggin had insisted on machine code compatibility between Intel's 8008 and 8080 microprocessors because he was concerned about providing an upgrade path for existing customers who wanted to migrate their code to newer microprocessors. He doesn't care much about instruction set compatibility of microcontrollers like the Z8. He argued that compatibility wasn't important for microcontrollers because customers usually didn't have legacy software to keep (at least not back then. But decades later, the situation is very different). Faggin envisioned an instruction set for the Z8 that would directly manipulate bits in the microcontroller's I/O ports, without requiring those bits to make a round trip through the microcontroller's accumulators or registers.

Although many online sources state that Zilog released the Z8 microcontroller in 1979, the company published two in-depth articles about the Z8 in the August 31, 1978 issue of Electronics magazine. Like many other early microcontrollers, the Z8 had a Harvard architecture with separate 16-bit program and data storage spaces. It also has a large 144-byte register file that can treat any register like an accumulator. Its instruction set and register set are different from those implemented in the Z80's more traditional von Neumann processor architecture.

Zilog has been through a lot of ups and downs as a company, but it's still in business and still selling microcontrollers based on the Z8 architecture, now called "Z8 Encore!" Z8 microcontrollers can be used in many low-cost peripheral applications , such as keyboard, mouse and infrared remote control. Some versions of the Z8 microcontroller now cost well under $1 each in high-volume production.

Faggin and Ungerman have long since left Zilog. Both became entrepreneurs. After leaving Zilog in 1980, Faggin founded or ran several companies, including Cygnet Technologies, Synaptics (an early pioneer of neural networks that eventually invented the touchpad), and Synaptics spin-off Foveon, which developed an Common multi-layer color image sensor. Ungermann left Zilog in 1979 and co-founded LAN pioneer Ungermann-Bass. IXYS acquired Zilog in 2009, and Littelfuse acquired IXYS in 2018 and owns Zilog.

Intel introduced the 8051, the successor to its 8048 microcontroller, in 1980. It became an immortal microcontroller, and it all started when an applications engineer forgot to bring his wallet to work one day and asked his boss at Intel to buy him lunch.

Intel released the 8048 microcontroller in 1976. The design's biggest weakness, limited memory addressability, emerged within the first year of product release. In a sense, this is a big problem because it shows that customers want more of the good stuff. Intel sold $7 million worth of its 8048 and 8748 microcontrollers in 1977 and expected sales to reach $70 million by 1980. In other words, the 8048 is popular.

On the other hand, the limited address space of the 8048 was baked into the architecture and instruction set. Bank switching bits in the registers double the size of the microcontroller's program and address space, but that's a problem to fix. Those who worked at Intel quickly learned that if the company wanted to capture a larger market for microcontrollers, it would need to improve the 8048 architecture. The improved architecture needs to be more tailored for future growth, rather than limitations built into the 8048 for expediency or cost reasons.

By 1977, John Wharton had been working as an applications engineer at Intel for about a year. He initially helped Intel customers design systems around the company's 8085 microprocessor, but soon became specialized in the design of 8048 microcontrollers, so Wharton became very familiar with all the shortcomings of the 8048 architecture.

One day in December 1977, Wharton went to work and realized he had forgotten his wallet. If he wants to have lunch, he has to find someone to buy him lunch. He went to his boss, Lionel Smith, and said, "I left my wallet at home, can you take me to lunch today?" Smith said he couldn't because he had scheduled a lunch meeting to discuss the 8048's architectural successor. However, Smith said, "They're always eating sandwiches there, and there's always leftover food, so why don't you come along and you just hide in the back and nibble on the leftovers?" Wharton agreed and attended the meeting, Because he was hungry and it was time for lunch. Wharton didn't know it at the time, but the deadline to decide on the 8048's replacement architecture was the end of the year, and the meeting he was about to attend would be critical to the 8048.

Wharton described the meeting in an oral history:

"I may not have gotten the details quite right because I'm more interested in the food and whether the potato salad will run out before it reaches the end of my table, etc. But they are talking about the various offshoots of the 8048. The low cost of the 8048 versions, low-power versions, ways to enhance the 8048 architecture, 16-bit machines that may be in development, that sort of thing.”

"With the 8048, the logical growth they identified seemed to be to expand the memory on the chip, but also expand some peripherals on the chip. The original [80]48 had a 1K of on-chip program memory, and about a year and a half later came an 8049, which had 2 KB of on-chip program memory and 128 bytes of RAM instead of 64 bytes of RAM."

"The logical next step is to crank it one more time and double the memory one more time, to 4K RAM, to 256 bytes of RAM, which will completely fill up the address space, and for this product, this will also be the production line "

"Because the 8048 was originally designed to solve the problem at hand, just to get everything onto one chip. It was an amazing product because it totally worked. But there was a mentality in that era that you were going to get Know what the hardware facilities are, and then almost as an afterthought come up with an instruction set that gives you full access to everything the chip designers have to offer.”

"So in order to extend the 8049 to the next chip, which logically should be called the 8050. The plan is to do some kind of modification of the 8048 instruction set and add bank switching instructions to increase the address space. Add I/O switching instructions to allow you Using a second timer instead of the first could do a lot to fill the product, but that seems to be the end of the line."

The next afternoon, Wharton had his customary one-on-one meeting with his boss. Smith asked, "So what did you think of yesterday's lunch?" Wharton said that if he were asked to design a successor to the 8048, the architectural changes being discussed were not the ones he would implement. "Why not?" Smith asked. Wharton replied:

"Well, because in the design work I do, and in conversations with clients and so on, the problems I have are not solved by this upgrade. If all you want to do is cram more functionality into an already In a somewhat tight package, you have to do that by removing things that are already there or making the product harder to use.”

What Wharton meant was that by the end of 1977, there were nearly ten variations of the 8048, and they were all a little different. In order to fit into the limited 8-bit instruction space of the 8048, instructions added to implement new functionality must replace instructions for deleted functionality in some microcontroller variants. Removed and added instructions make the 8048 variants somewhat incompatible, which complicates code portability and makes it difficult to change designs from one 8048 variant to another.

This made Smith realize that Wharton, who was sitting across from him, had the opportunity to be the right person to define the next generation of Intel's microcontroller architecture. He asked Wharton to develop a new microcontroller architecture that would overcome the 8048's shortcomings. Wharton spent three days from Friday to the weekend to develop and deliver the architectural plan the following Monday. After much discussion of Intel's famous "constructive confrontation" but little substantive change, Wharton's architecture essentially became the Intel 8051 - the microcontroller that never dies.

Intel began sampling the 8051 microcontroller in 1980. One of the key differences between the 8051 and the 8048 is that in-circuit emulation has become important due to the increasing size of the processor program space and the increasing complexity of the target applications. Intel manufactured a bonded version of the 8051 called the 8051E , which provides the internal address and data buses and control signals required to develop an in-circuit emulator.

Furthermore, the basic chip layout of the 8051 was designed from the beginning to easily fit ROM or EPROM into the space reserved for program memory. The EPROM cells were much larger than the ROM cells, so one side of the 8051's physical layout had to be pushed out to make room for the EPROM, but this strategy proved very effective and could get all the parts out of the factory in a short time.

The 8051 proved to be a milestone for Intel, with unit shipments climbing into the billions per year, and Intel sold 8051 microcontrollers for decades. In 1998, Wharton conducted a survey of semiconductor suppliers using the 8051 architecture and found that there were five major suppliers offering microcontrollers based on 8051 designs. In total, these suppliers provided more than 200 device variants. In 2006, Wharton attended an embedded systems conference in California and picked up a flyer from Keil Software, which provides software development tools for the 8051. The flyer lists over 60 companies offering over 1000 different 8051 variations. By any measure, the 8051 microcontroller was a huge success, continuing to sell well for a quarter of a century after its introduction.

In the 8051's oral history, Wharton explains the 8051's longevity this way:

“In the embedded control market, what we are doing is controlling the world, interacting with the world, interacting with humans, interacting with machines, turning motors on and off, gasoline pumps, cash registers, keyboards, cell phones, digital cameras, and in these markets, What you're doing is a process of primarily controlling a project, looking at inputs, making decisions, controlling outputs, but you're doing it at real-world speeds, and the real-world hasn't changed much at the speed that people typed 30 years ago. Pretty much, so if a typewriter from 30 years ago is enough, it still works."

John Wharton passed away in 2018, but his most successful creation lives on, proving that when products are defined by experienced and observant application engineers rather than by "experts" working in factories, processor architecture can Big difference. You'll still find 8051 microcontrollers in current products, from packaged microcontroller chips in computer mice to microcontroller IP cores integrated into Bluetooth chips. Many real world microcontroller applications do not require the microcontroller to have more features than the 8051. They didn't have it in 1980 and they still don't have it now.

I wasn't going to write this chapter in my series on the history of early microcontrollers, not because the PIC1650 microcontroller wasn't important, but because it wasn't very important in its original incarnations. General Instrument Microelectronics (GI) envisions the PIC1650 microcontroller as a peripheral chip for its 16-bit CP1600 microprocessor. In fact, "PIC" originally stood for "peripheral interface chip," but the company eventually changed that meaning to "programmable intelligent computer." Neither the CP1600 microprocessor nor the PIC1650 microcontroller had much impact on the industry in the 1970s. However, GI endured several near-death experiences and was reborn as Microchip in the late 1980s.

GI was early to the MOS LSI party—in fact, the company introduced the first commercial MOS IC, a 20-bit shift register, in 1964—and the company built a large and successful portfolio of MOS LSI chips. These include but are not limited to:

• early calculator chips

• Clocks and Clocks - Radio Chips and Modules

• frequency counter chip

• TV tuners and TV game chips

• Music chips and sound generators

• text-to-speech synthesizer

• digital voltmeter chip

• ROM, SRAM, EAROM (electrically variable ROM)

• keyboard controller

• Communication chips including UART (Universal Asynchronous Receiver/Transmitter), P/SAR and P/SAT (Programmable Synchronous/Asynchronous Receiver and Transmitter)

This is a huge product line.

GI partnered with Honeywell to develop a 16-bit microprocessor, launching a device called the CP1600 in 1975. It is loosely based on Digital Equipment Corporation's PDP-11 minicomputer architecture. Honeywell uses this microprocessor in its process control computers (called multifunction controllers) and related equipment. Mattel Electronics used a version of the CP1600, called the CP1610, in the Intellivision home video game console unveiled at CES in January 1979. Apparently, the CP1600 isn't used in many products other than Honeywell's equipment and Mattel Intellivision.

The CP1600 disappeared from GI catalogs in the early 1980s. The CP1610 survived for a few more years. Adam Osborne, the great microprocessor chronicler of the 1970s, wrote that the main reason for the CP1600's lack of Design Win was the lack of support from GI, which tended to focus its efforts on supporting high-volume customers like Mattel.

Interestingly, Honeywell multifunction controllers from that era are still being refurbished and supplied by companies such as Western Process Computers, which supplied products manufactured by Azbil (formerly Yamatake, then Honeywell's process control partnership in Japan). FPGA-based CP1600 simulation developed by partners. This CP1600 simulation serves as a support strategy to keep these older process controllers operational after the eventual failure of the 40-year-old GI microprocessors. This example gives a hint of how long some of this old equipment may continue to be used.

The CP1600 microprocessor uses memory-mapped I/O, but GI did not develop conventional peripheral chips for the microprocessor's bus. Instead, GI developed an I/O microcontroller called the PIC1650 to extend the I/O capabilities of the CP1600. GI launched the PIC1650 microcontroller in 1976, and Intel announced the 8048 in the same year.

The PIC1650 was not the first microcontroller, nor was it the best. Like most microcontrollers, the PIC1650 comes in a 40-pin DIP package, runs at 5 volts, and uses a quirky Harvard architecture with 12-bit instruction words, 8-bit data words, and a 2-level stack. The PIC1650's only on-chip RAM is a 32-byte register file. The original PIC1650 had a relatively small 512 words of ROM for program storage.

For several years now, GI has not offered a PIC1650 with UV-erasable EPROM or EAROM (EEPROM, to use more modern terminology). Instead, the company developed the PIC1664, which brought out the address and data buses, thus requiring a 64-pin package. In comparison, the PIC1650's competitors have larger ROM, larger RAM, and are available in either a UV-erasable EPROM version with a window package or with a piggyback socket that accepts UV-erasable EPROM. By 1980, GI added a 3-level stack and a "return" instruction for subroutine calls to the PIC1656.

Brian Harden was working at GI's Glenrothes factory in Fife, Scotland, when the PIC1650 was introduced in the mid-1970s. GI Glenrothes received a TTL emulation board for the microcontroller in the early days of the PIC1650, and Harden was developing a teletext/viewdata chip for GI at the time. He needed some way to implement a user interface for the chip and decided to try using a microcontroller emulator to implement the user interface. He succeeded, setting a precedent for the company. Over the next few years, GI developed a variety of preprogrammed devices based on its PIC1650 microcontroller.

Harden liked microcontrollers and continued to use them after leaving GI in 1981. He founded a design consulting company specializing in PIC1650-based designs. One of his early designs was the control electronics used in Creda Micron washing machines. He replaced the mechanical controls with PIC microcontrollers. Creda Micron integrates multiple PIC microcontrollers for motor speed control, user interface and overall machine management. Harden ran his design consulting firm until 1998 and estimates that three-quarters of his designs were based on PIC microcontrollers.

By 1981, GI added the PIC1670, doubling the amount of on-chip ROM. A slight architectural extension provides the PIC1670 microcontroller with a 13-bit instruction word instead of the PIC1650's 12-bit instruction word, which allows the PIC1670 to directly address a larger on-chip register file and allows for some additional instructions. By 1983, there were seven members of the PIC1650 family, including the first CMOS member, the PIC16C58 - a low-power version of the PIC1655A - which was the PIC1650A in a low-cost 28-pin package. Even then, GI's goal was to use PIC microcontrollers for the most cost-sensitive designs.

General Instrument spun off its microelectronics division into a wholly owned subsidiary in 1987. A group of venture capitalists purchased the subsidiary and named it Microchip Technology in 1989. However, things didn't go smoothly for Microchip. Its sales were flat or declining, mainly due to competition from Japanese semiconductor companies. More than 60% of the company's revenue comes from the disk drive industry, and one customer accounts for more than 25% of the company's annual revenue. Microchip had an aging factory with poor quality control, resulting in low yields and high manufacturing costs because of excessive scrap. Incredibly, the company sells and ships some of its products below cost. By 1990, the company was losing $2.5 million per quarter and had only enough cash on hand for six months. Microchip is in a weak position. What the company needs is focus.

Steve Sanghi joined Microchip on February 28, 1990 as executive vice president. That day, as he was signing employment documents in the human resources department, he learned over the company intercom that the current CEO would be fired and that the board of directors was beginning a search for a suitable candidate, a new CEO. Sanghi was asked to come to the board from HR, where he was asked to leave the company as part of a management reshuffle. He didn't leave. Three months later, he became Microchip's new CEO. For the next nine months, Microchip teetered on the edge of bankruptcy while Sanghi worked to turn things around.

Sanghi started a product line triage to find profitable devices that Microchip could produce through its factories, and the winners included the PIC1650 architecture, UV-erasable EPROM and EEPROM. By combining these three on-hand technologies, Microchip assembled the first low-cost, field-programmable PIC microcontrollers that cost just a few dollars each, while competing microcontrollers from Motorola and Intel sold for just a few dollars each. The price is over $10. Field-programmable PIC microcontrollers are very attractive to distributors because they can be placed on the shelf and ready to be shipped to customers. Distribution has become an important channel for Microchip's PIC microcontrollers.

The first Microchip PIC microcontroller was so successful that the company quickly grew the product line to include 16-bit and 32-bit varieties. In 1990, Microchip was ranked as the 20th largest microcontroller company. By 2002, it had become the number one microcontroller supplier. During those years, Microchip targeted just one company: Motorola, the leader in microcontrollers. The Motorola MC6800 microprocessor architecture has evolved into numerous microcontroller families, including the MC6801, MC6805, MC68HC05, MC68HC08, MC68HC11 and MC68HC12. These vast families put Motorola at the top of the list of microcontrollers. Microchip managed to push the PIC1650 architecture from the bottom of the list to the top within a decade.

In a subsequent development, on-chip EEPROM technology allowed Microchip to place its erasable microcontrollers in low-cost plastic packages, while other microcontroller vendors were forced to use expensive windowing for their UV-erasable parts. Ceramic package. Microchip supports its customers well, and its inexpensive development platforms echo and amplify the appeal of its low-cost microcontrollers. Microchip also employs an active academic program to teach budding developers how to use Microchip microcontrollers instead of competing products. The program sowed the seeds for Microchip's future markets. By 2002, all of these strategies helped Microchip rise to the top of the microcontroller market share.

Microchip's beginnings occurred early in the history of microcontrollers, but the company's success occurred decades later, during an era when microcontroller usage began to explode and new microcontroller families began to appear at a very rapid rate. During this period of expansion, European and Japanese semiconductor manufacturers rushed into the market.

In this article, I have not covered microcontrollers from the 1980s onwards, nor have I covered all microcontrollers developed in the 1970s. I know I've omitted some microcontrollers because of a lack of documented information other than the datasheet, which makes for a poor history. For example, I didn't find enough information to write a full article on National Semiconductor's COP microcontroller family.

一些单独的微控制器历史在网上或计算机历史博物馆和斯坦福大学图书馆等机构的口述历史记录中得到了很好的介绍。我甚至在旧电子邮件列表服务器和在线公告板的在线档案中找到了很多有用的资料，而且我在微控制器供应商自己的未经审核的在线用户对用户论坛中发现了一些隐藏的项目。其他不太流行或可能不太出名的微控制器的文档很少，甚至完全消失在时间的迷雾中。

I hope this series of articles can give everyone an understanding of the early history of the MCU.

*Disclaimer: This article is original by the author. The content of the article is the personal opinion of the author. The reprinting by Semiconductor Industry Watch is only to convey a different point of view. It does not mean that Semiconductor Industry Watch agrees or supports the view. If you have any objections, please contact Semiconductor Industry Watch.

Today is the 3274th content shared by "Semiconductor Industry Observation" with you. Welcome to pay attention.