Chip Hall of Fame: Intel 4004 Microprocessor

The first CPU-on-a-chip was a shoestring crash project

Intel 4004

The Intel 4004 was the world’s first microprocessor—a complete general-purpose CPU on a single chip. Released in March 1971, and using cutting-edge silicon-gate technology, the 4004 marked the beginning of Intel’s rise to global dominance in the processor industry. So you might imagine that the full resources of Intel—still a fledgeling company at the time—were devoted to this groundbreaking project. But in fact, the 4004 was an understaffed side project, a crash job that nearly crashed, one simply intended to drum up some cash while Intel developed its real product line, memory chips.

As described by Ken Shirrif in a July 2016 feature for IEEE Spectrum, the increasing transistor count and complexity of integrated circuits in the 1960s meant that by 1970, multiple organizations were hot on the path to the microprocessor. Some of these, like Texas Instruments, had a lot more resources than Intel. So why did Intel, founded just a few years earlier, in 1968, cross the finish line first? It was largely thanks to four engineers, one of whom didn’t even work for the company. (For a lengthy version of this story from the engineers themselves, you can read their oral history panel, as captured by the Computer History Museum).

The first of the four engineers is Masatoshi Shima, who worked for Japanese office calculator company Busicom, which wanted to create a newly computerized calculator. In April 1969, Busicom and Intel signed a provisional agreement for Intel to develop a custom set of chips for the calculator. Consequently, in June 1969 Shima and others travelled to Intel to discuss the plans in more detail. Shima proposed an eight-chip system: three chips to interface with peripherals such as the keyboard and printer, one chip to store data, one chip to store program code, and two chips that together would make up the CPU.


Masatoshi Shima at the Computer History Museum’s 2009 Fellows Award event, and the Busicom calculator that was the target application for the world’s first microprocessor.

Ted Hoff is the second engineer in our tale and was the head of the Intel applications department that was negotiating with Busicom. Hoff was worried that Intel would struggle to produce so many chips, especially because the system would require many pins per chip to interconnect, which would push the limits of the ceramic packing technology Intel was using. He proposed halving the chip count: one 256-byte program memory chip, dubbed the 4001, one 40-byte data memory chip, the 4002, a peripheral interface chip, the 4003, and one CPU chip, the 4004. The whole system—called the MCS-4—would be 4-bit, significantly reducing the number of pins needed to interconnect the chips. Hoff brought in engineer No. 3, Intel’s Stanley Mazor. Together Hoff and Mazor put together a set of specs for each chip and a proposed production schedule.

At a follow-up meeting in October of 1969, Intel made its counterproposal. Busicom was interested and Shima returned to Japan to prototype software for the new calculator to make sure the MCS-4 architecture would support Busicom’s needs. An agreement was made in February 1970, with Busicom planning its calculator rollout on the basis of Hoff’s and Mazor’s schedule. It was decided that Shima would come back to California to check on progress in April 1970. The chips were to be put into production on a staggered schedule from July to October 1970, starting with the 4001 and ending with the 4004.

However, unbeknownst to Shima and Busicom, the 4004 project had ground to a halt inside Intel in early 1970. The problem was that Hoff and Mazor were not chip designers—those people who can take specifications and create detailed logic-gate diagrams. Those diagrams, in turn, are used to work out exactly how and where transistors and other components are to be patterned on the physical chip.

In fact, there was no one at Intel who could take on the job, as the company was then focused on developing memory chips. Finally, Intel made one of the great hires of all time and introduced the fourth critical person in this story: Frederico Faggin, a young engineer uniquely suited to the job. At the start of his career, Faggin had designed and built a computer from scratch for Olivetti, in Italy. Then in the late 1960s, he had joined Fairchild Semiconductor, in Silicon Valley, where he made key contributions to the advanced metal oxide semiconductor (MOS) technology that Intel’s chips relied on. Faggin wanted to work in a more entrepreneurial environment than Fairchild, and so accepted an offer from Intel in April 1970.

On Faggin’s first day on the job, Mazor briefed him on the Busicom project. As Faggin wrote in his personal account of the 4004’s development for the Winter 2009 issue of IEEE Solid-State Circuits Magazine, when he saw the schedule: “My jaw dropped: I had less than six months to design four chips, one of which, the CPU, was at the boundary of what was possible.”


From left, Federico Faggin, Ted Hoff, and Stanley Mazor holding Intel 4004 processors at the National Inventors Hall of Fame in 1996

The original schedules were based on estimates suitable for designing memory chips—which use many repeating elements—rather than processor chips, which use complex and varied logic circuits. In addition, Faggin had no support staff and none of the tools and infrastructure that other companies had to help create and test digital logic designs.

A few days after Faggin’s start, Shima landed in the States for his progress check. Mazor and Faggin went to pick him up from the airport and bring him back to Intel. Shima was expecting to see a logic-level plan for the chips that he could check against the agreed-upon specifications. “Shima was furious when he found out that no work had been done in the five months and he became very angry at me…. It took almost one week for Shima to calm down,” wrote Faggin.

Faggin worked out a new schedule, and it was agreed that while Intel set about hiring more people for the project, Shima would stay for six months to help with the design. Faggin himself dived into 70- to 80-hour workweeks.

Faggin worked through the chips in order of complexity: the 4001 ROM, followed by the 4003 interface chip, then the 4002 RAM, followed finally by the 4004 CPU. Shima checked the logic of the chips and provided feedback on how they would fit into Busicom’s larger calculator design. At the end of 1970, the chip design was complete. Faggin added a personal flourish to the CPU’s layout: He placed his initials along the edge of the processor, a microscopic “F.F.” etched into every 4004 made. Busicom finally had a complete working set of MCS-4 chips in March 1971.


Frederico Faggin and an enlarged picture of the Intel 4004 die. The 4004 had 2,300 transistors.

As Busicom had commissioned the chipset, it had exclusive rights to the design, preventing Intel from selling the 4004 to anyone else. But after some prompting from Hoff and others about the processor’s potential, Intel offered to give Busicom a break on the cost of the chips if Intel could sell the 4000 families for noncalculator applications. Busicom agreed, and Intel began advertising the 4004 in November 1971: “Announcing a new era of integrated electronics,” blared the ad copy—a rare case of absolute truth in advertising.

How to Fly a Drone With Your Body

For real and simulated drones, piloting with torso movements outperforms a joystick every time—and it’s easier to learn – Megan Scudellari

A model demonstrates a body-machine interface for controlling a simulated drone.

Using only the movements of one’s torso to pilot a drone is more intuitive—and more precise—than a joystick, according to new research from engineers at the École Polytechnique Fédérale De Lausanne (EPFL) in Switzerland.

The technique, tested in virtual reality and with real drones, requires less mental focus from the pilot and frees up their head and limbs. So, for instance, a drone operator at a natural disaster site or on a search and rescue mission could concentrate on looking around and analyzing visual information rather than controlling the flight path of the drone.

The team also found that torso control is easier to learn and more intuitive than a traditional joystick for most people. “It’s not that a joystick does not work—pilots for drone racing do amazing things with their joysticks—but we’ve noticed that for some people, it can be difficult to learn and you have to be really focused while you’re doing it,” says study author Jenifer Miehlbradt, a graduate student at EPFL.

In a series of experiments described this week in the journal PNAS, a team led by Miehlbradt and EPFL neuro-engineer Silvestro Micera set out to come up with an alternative, easier way to pilot a drone.

Infrared markers on a volunteer

First, they stuck over a dozen infrared markers all over the upper body of 17 volunteers and asked them to follow a virtual drone through a simulated landscape in virtual reality. “We asked them to follow the movements of the drone with their body in a way that felt natural to them,” says Miehlbradt. One participant opted to fly the drone-like Superman—with one arm extended above his head—and another chose to “swim” through the air, but everyone else used either their torso alone or their torso and arms to glide like a bird.

Next, in a first-person virtual reality simulation, 39 volunteers were asked to follow a path of clouds as closely as possible. Across the board, torso control was easier to learn and more precise than torso and arms or joystick control. Plus, it actually feels like flying, says Miehlbradt. Finally, it was time to try out the torso technique with real drones. Participants were allowed to train for nine minutes in virtual reality, they were given control of a quadcopter with FPV video feedback and allowed to freely fly for two minutes to get used to its dynamics. “At first, it’s a bit scary,” says Miehlbradt. “It takes a minute to get used to this feeling of ‘I’m over there, with this object that is moving.’ It’s extremely immersive.”

In their final test, volunteers were asked to steer the drone through six gates arranged along a figure-eight trajectory. With the aforementioned minimal training, they did well, steering the quadcopter through the gates without collisions 88 percent of the time. These initial experiments were done with reflective markers on the body and a motion-capture system involving cameras set up around the subject. While a tried-and-true method for motion analysis, such a system is too bulky and expensive for widespread, commercial use.

Now, the second team at EPFL has built the “FlyJacket”—a soft jacket with a motion-sensing device on the back, an arm-support system to prevent fatigue, and VR goggles for simulation. This portable system could eventually be applicable to consumer drones or other types of robots. In the future, the team’s screening method could also be used to identify common, intuitive control patterns for robots of various shapes, says Miehlbradt. Maybe even a flying robot that can transform its shape in mid-air?

And, yes, we know you’re thinking it: This type of body control could—and very likely will—be applied to virtual reality and other types of gaming. During development, the team often set up demonstrations on campus to let people try out flying the drones. The response was unequivocal: “They love it,” says Miehlbradt with a laugh. “It’s something new. It really gives you a feeling of flying…I think it could become more popular than a joystick.”


Swarm Asks FCC for Permission to Have Its Rogue Satellites Phone Home

A new application seeks approval to beam the company’s unlicensed CubeSats’ location, course, and speed down to Earth—and then share that data online

Illustration of Swarm's cube sats trying to contact home

Ever since Swarm Technologies illegally launched four experimental satellites in January, the tiny spacecraft has been orbiting the Earth in silence, awaiting judgement from a furious U.S. Federal Communications Commission (FCC). Now the stealthy startup wants to briefly activate the satellites, probably with the aim of placating the FCC and smoothing the way for future launches.

new FCC application, filed late last week, is seeking authority to legally transmit data from the SpaceBEE satellites for the first time. However, Swarm does not want to test its innovative Internet of Things communications technology, which the company believes could be orders of magnitude cheaper than existing satellite links. Instead, the transmissions would comprise only “telemetry packets containing essential orbital data. This data includes rotational and magnetometer data, as well as comprehensive GPS tracking information, such as latitude/longitude, altitude, course, speed, and satellite count.”

That data would then be made available to the FCC and shared with other agencies and satellite operators via Swarm’s website (although this link is currently unavailable). The FCC originally denied Swarm’s application for the four CubeSats because it believed that, at just 10 centimetres (cm) x 10 cm x 2.8 cm, they were too small to be tracked reliably on orbit. Uncertainty about their positions would thus expose other satellites and potentially even crewed spacecraft to unacceptable collision risks.

Swarm thought that it had mitigated those risks by including a GPS unit in each satellite to report its position and integrating passive reflectors to boost the tiny satellites’ visibility to ground-based radar stations. The FCC disagreed and told Swarm not to launch, but the company did so anyway, aboard an Indian commercial rocket.

When the FCC learned of the unlicensed launch, it set aside Swarm’s authorization for a subsequent launch of larger satellites, delayed permission for a market trial of the Internet of Things system, and started an investigation into the company.

Curt Blake, president of Spaceflight Industries, the company that manifested the SpaceBEEs on the Indian rocket, participated in that investigation. “We went before the FCC and told them what had happened,” he told the NewSpace 2018 conference in Seattle in June. “It’s fair to say [Swarm] misrepresented to us what their status was with the government in terms of [their] regulatory status.” Blake also said the FCC was working on a new regulatory policy, and that “without question” Spaceflight would not put an unlicensed satellite on a launch vehicle again.

Six months after the Swarm launch, the only public output from the FCC investigation has been an advisory notice warning satellite operators that ignoring its rules “can and will result in enforcement action.” That typically involves a large fine.

Meanwhile, Swarm has been feeling increasing pressure. In early June, it filed another application for permission to launch three experimental satellites, writing that it “urgently needs to demonstrate the viability of its proposed satellite-based communications network to technical and business partners, potential investors, and potential customers.” Then, less than two weeks ago, lawyers for Swarm requested the voluntary dismissal of two pending and set-aside applications at the FCC. This was necessary, they wrote, “given the need for Swarm to realign experiments and/or market trials with currently available launch vehicles.” Basically, Swarm had missed a ride to orbit while the FCC was deliberating, and is now focusing its efforts on a SpaceX rocket due to launch from Vandenberg Air Force Base in September or October.

Swarm's listing of orbital parameters from the FCC

Neither Swarm nor the FCC immediately responded to requests for comment on the latest filing, but there are two likely explanations for it. One is that the FCC is requiring the company to make the SpaceBEEs’ orbital data widely available as part of an as yet unannounced enforcement action. The second is that Swarm is proactively offering that data in an effort to placate the FCC, and possibly accelerate its decision-making regarding the upcoming mission.

The filing says that the sole purpose of the application is to collect orbital and tracking data. This would involve a handful of 2-second transmissions from each satellite each day, as they pass over ground stations in Silicon Valley and the state of Georgia. All the data will then be uploaded, within half a second, to a portal on Swarm’s website, where it could be used by federal agencies, satellite tracking companies and other operators.

Although Swarm is proposing this somewhat convoluted process, it still believes that its SpaceBEEs are no more dangerous than any other CubeSat on orbit. In the new filing, it notes: “All four SPACEBEE satellite [sic] have been trackable by the Space Surveillance Network (SSN) by normal means. No gaps in tracking have occurred, and the satellites are currently being tracked, and the… orbital data is being posted publicly to the SSN database.”

The four SpaceBEEs do all have live entries in the U.S. Air Force’s Space Command catalogue. If the new data were essential for orbital safety, Swarm would have to file yet another application in six months anyway.

DARPA Wants Your Insect-Scale Robots for a Micro-Olympics

SHRIMP is a new DARPA program to develop insect-scale robots for disaster recovery and high-risk environments- Evan Ackerman

DARPA's SHRIMP program wants to develop insect-scale robots for disaster recovery and high-risk environments

The DARPA Robotics Challenge was a showcase for how very large, very expensive robots could potentially be useful in disaster recovery and high-risk environments. Humanoids are particularly capable in some very specific situations, but the rest of the time, they’re probably overkilling, and using smaller, cheaper, more specialized robots is much more efficient. This is especially true when you’re concerned with data collection as opposed to manipulation—for the “search” part of “search and rescue,” for example, you’re better off with lots of very small robots covering as much ground as possible.

Yesterday, DARPA announced a new program called SHRIMP: SHort-Range Independent Microrobotic Platforms. The goal is “to develop and demonstrate multi-functional micro-to-milli robotic platforms for use in natural and critical disaster scenarios.” To enable robots that are both tiny and useful, SHRIMP will support fundamental research in the component parts that are the most difficult to engineer, including actuators, mobility systems, and power storage.

From the DARPA program announcement:

Imagine a natural disaster scenario, such as an earthquake, that inflicts widespread damage to buildings and structures, critical utilities and infrastructure, and threatens human safety. Having the ability to navigate the rubble and enter highly unstable areas could prove invaluable to saving lives or detecting additional hazards among the wreckage. Partnering rescue personnel with robots to evaluate high-risk scenarios and environments can help increase the likelihood of successful search and recovery efforts, or other critical tasks while minimizing the threat to human teams.

Technological advances in microelectromechanical systems (MEMS), additive manufacturing, piezoelectric actuators, and low-power sensors have allowed researchers to expand into the realm of micro-to-milli robotics. However, due to the technical obstacles experienced as the technology shrinks, these platforms lack the power, navigation, and control to accomplish complex tasks proficiently.

To help overcome the challenges of creating extremely SWaP-constrained microrobotics, DARPA is launching a new program called SHort-Range Independent Microrobotic Platforms (SHRIMP). The goal of SHRIMP is to develop and demonstrate multi-functional micro-to-milli robotic platforms for use in natural and critical disaster scenarios. To achieve this mission, SHRIMP will explore fundamental research in actuator materials and mechanisms as well as power storage components, both of which are necessary to create the strength, dexterity, and independence of functional micro-robotics platforms.

That term “SWaP” translates into “size, weight, and power,” which are just some of the constraints that very small robots operate under. Power is probably the biggest one—tiny robots that aren’t tethered either run out of power within just a minute or two or rely on some kind of nifty and exotic source, like lasers or magnets. There’s also control to consider, with truly tiny robots almost always using off-board processors. These sorts of things substantially limit the real-world usefulness of microrobots, which is why DARPA is tackling them directly with SHRIMP.

One of our favourite things about DARPA programs like these is their competitive nature, and SHRIMP is no exception. Both components and integrated robots will compete in “a series of Olympic-themed competitions [for] multi-functional mm-to-cm scale robotic platforms,” performing tasks “associated with manoeuvrability, dexterity, [and] manipulation.” DARPA will be splitting the competition into two parts: one for actuators and power sources, and the other for complete robots.

Here are the tentative events for the actuator and power source competition; DARPA expects that teams will develop systems that weigh less than one gram and fit into one cubic centimetre.

High Jump: The micro-robotic actuator-power system must propel itself vertically from a stationary starting position, with distance measured only in the vertical direction and survivability as the judging criteria. Expected result: >5cm.

Long Jump: The micro-robotic actuator-power system must propel itself horizontally from a stationary starting position, with the distance measured only in the horizontal direction and survivability as the judging criteria. Expected result: >5cm

Weightlifting: The micro-robotic actuator-power system must lift a mass, with progressively larger masses until the actuator system fails to lift the weight. Expected result: >10g. 

Shotput: The micro-robotic actuator-power system must propel a mass horizontally, with the distance measured only in the horizontal direction as the judging criteria. Both 1-gram and 5-gram masses must be attempted. Expected result: >10cm @ 1g, >5cm @ 2g.

Tug of War: The micro-robotic actuator-power system will be connected to a load cell to measure the blocking force of the actuator mechanism. Expected result: > 25mN.

Teams competing with entire robots will have a separate set of events, and DARPA is looking for a lot of capability in a very, very small package—in a volume of less than one cubic centimeter and a weight of less than one gram, DARPA wants to see “a micro power source, power converters, actuation mechanism and mechanical transmission and structural elements, computational control, sensors for stability and control, and any necessary sensors and actuators required to improve the maneuverability and dexterity of the platforms.” The robots should be able to move for 3 minutes, with a cost of transport of less than 50. Teams are allowed to develop different robots for different events, but DARPA is hoping that the winning design will be able to compete in at least four events.

Rock Piling: For each attempt, the microrobot must travel to, lift, and stack weights (varying from 0.5 to 2.0 g) in a minimum of two layers without human interaction. Expected result: 2g, 2 layers. 

Steeplechase: Competing teams will be given precise locations and types of obstacles (e.g. hurdle, gap, step, etc.) relative to the starting location. For each attempt, the microrobot must traverse the course without human interaction or recharge between each obstacle. The number of cleared obstacles and total distance will be used as the judging criteria. Expected result: 2 obstacles, 5m.

Biathlon: Competing teams will be given the choice between three beacon types (temperature, light, or sound) or they may choose to use all 3 types of beacons. For each attempt, the microrobot must traverse to a series of beacon waypoints to create an open circuit without human interaction or recharge between each waypoint. Expected result: 2 beacons, 5m.

Vertical Ascent: Microrobots will traverse up two surfaces, one with a shallow incline (10º) and the other with a sharp incline (80º). The total vertical distance travelled will be the judging criteria. Expected result: 10m at 10°, 1m at 80°.

DARPA has the US $32 million of funding to spread around across multiple projects for SHRIMP. Abstracts are due August 10, proposals are due September 26, and the competition could happen as early as March of next year.

Intelligent Machines A team of AI algorithms just crushed humans in a complex computer game- Will Knight

Algorithms capable of collaboration and teamwork can outmanoeuvre human teams.

Five different AI algorithms have teamed up to kick human butt in Dota 2, a popular strategy computer game.

Researchers at OpenAI, a nonprofit based in California, developed the algorithmic A team, which they call the OpenAI Five. Each algorithm uses a neural network to learn not only how to play the game, but also how to cooperate with its AI teammates. It has started defeating amateur Dota 2 players in testing, OpenAI says.

This is an important and novel direction for AI since algorithms typically operate independently. Approaches that help algorithms cooperate with each other could prove important for commercial uses of the technology. AI algorithms could, for instance, team up to outmanoeuvre opponents in online trading or ad bidding. Collaborative algorithms might also cooperate with humans.

OpenAI previously demonstrated an algorithm capable of competing against top humans at single-player Dota 2. The latest work builds on this using similar algorithms modified to value both individual and team success. The algorithms do not communicate directly except through gameplay.

“What we’ve seen implies that coordination and collaboration can emerge very naturally out of the incentives,” says Greg Brockman, one of the founders of OpenAI, which aims to develop artificial intelligence openly and in a way that benefits humanity. He adds that the team has tried substituting a human player for one of the algorithms and found this to work very well. “He described himself as feeling very well supported,” Brockman says.

Dota 2 is a complex strategy game in which teams of five players compete to control a structure within a sprawling landscape. Players have different strengths, weaknesses, and roles, and the game involves collecting items and planning attacks, as well as engaging in real-time combat.

Pitting AI programs against computer games has become a familiar means of measuring progress. DeepMind, a subsidiary of Alphabet, famously developed a program capable of learning to play the notoriously complex and subtle board game Go with superhuman skill. A related program then taught itself from scratch to master Go and then chess simply by playing against itself.

The strategies required for Dota 2 are more defined than in chess or Go, but the game is still difficult to master. It is also challenging for a machine because it isn’t always possible to see what your opponents are up to and because teamwork is required.

The OpenAI Five learn by playing against various versions of themselves. Over time, the programs developed strategies much like the ones humans use—figuring out ways to acquiring gold by “farming” it, for instance, as well as adopting a particularly strategic role or “lane” within the game.

AI experts say the achievement is significant. “Dota 2 is an extremely complicated game, so even beating strong amateurs is truly impressive,” says Noam Brown, a researcher at Carnegie Mellon University in Pittsburgh. “In particular, dealing with hidden information in a game as large as Dota 2 is a major challenge.”

Brown previously worked on an algorithm capable of playing poker, another imperfect-information game, with superhuman skill (see “Why poker is a big deal in AI”). If the OpenAI Five team can consistently beat humans, Brown says, that would be a major achievement in AI. However, he notes that given enough time, humans might be able to figure out weaknesses in the AI team’s playing style.

Other games could also push AI further, Brown says. “The next major challenge would be games involving communication, like Diplomacy or Settlers of Catan, where balancing between cooperation and competition is vital to success.”