Can Cheap Computing Put the Internet Everywhere?

It’s in your car. It’s in your home. Soon, it may monitor your digestive system.

As our lives become increasingly surrounded by devices and objects that can communicate, collect, and transmit data, we’re brushing up against the advent of the so-called Internet of Things. Known as IoT in tech-speak, it promises to connect our refrigerators with our toasters, our lights with our thermostats, and our fitness monitors back with our refrigerators.

Small, inexpensive chips like this can work with the near-field communication (NFC) standard to transmit data to or read information from other devices like phones.

Experts have high hopes for IoT. The consultancy McKinsey predicts it will drive $3.9–11.1 trillion of economic activity by 2025. For IoT to have that impact, though, it’ll have to be more than connected refrigerators and smart thermostats. The sensors and computers that will drive IoT will have to be in everything from streetlights to heartrate monitors to bottles of wine.

To enable this kind of interconnectivity, computer chips will need to get cheap, fast. And they’ll need a reason for being. Up until now, IoT devices have been neither sufficiently ubiquitous or necessary to succeed.

A company called ThinFilm hopes to change that with the opening of a new semiconductor plant in San Jose, California. ThinFilm thinks their new process could drop prices enough to usher in the IoT dream. While the plant still hosts the iconic “bunny suits” and enclosed rooms, ThinFilm is partially printing chips rather than fully etching them as most chips are today. The new process has been mooted and tested in labs for years, and a commercial version could produce chips that are significantly cheaper. Moving to roll-to-roll production could allow ThinFilm to make billions of these tiny devices, and it could allow a technology known as near-field communication, or NFC, to become ubiquitous.

If ThinFilm’s roll printing process works, they could get there. It’s a big if, but the payoff could be big, both for the company and for the people who may need cheap chips. “New technology has to somehow introduce a new level of convenience at a new low cost,” says Woodward Yang, a professor of electrical engineering and computer science at Harvard University. “People don’t buy technology, people buy solutions.”

Connecting Devices

Networks of everyday devices predate the concept of IoT by many decades. ATMs, one of the first examples of a connected device, were used as early at 1974. More recently, as chip costs have gone down, intelligent connected devices have become more common, starting with computers and moving into phones, speakers, and even headphones.

In 2008, networking company Cisco estimated that there were more objects connected to the internet than humans on Earth. By 2020, analysts at the market research group Gartner predict that over 26 billion devices will be connected. But in order for IoT to truly take off, it needs to be more than just a simple chip in a card that unlocks your hotel room or scans you into your workplace.

Cheap chips could push connected devices a bit further into our lives. It’s a problem that ThinFilm thinks it can fix. The new plant is currently producing chips on sheets of stainless steel, printing a portion of the devices instead of exclusively using traditional semiconductor fabrication methods that require many time-consuming steps and expensive equipment. Ultimately, they expect that partially printing chips will increase production volume.

The most visible use of NFC is payment systems like Apple Pay and Samsung Pay.

The chips that ThinFilm is making are not the first ones in this space. Barcodes are cheap to print, and quick to read, operating as basic ID codes. Radio frequency identification (RFID) chips provide more information and can be updated—they work at broader ranges than static barcodes. ThinFilm is focusing on near-field communication (NFC) devices such as the ones that enable ApplePay and SamsungPay. These chips allow to devices to communicate at close ranges, usually within about 1.5–8 inches. NFC can operate more quickly and consume less energy than the connections which currently wirelessly connect your earbuds to your phone. RFID chips typically require a memory chip tag, reader, and antenna, while NFC has the reader and tag integrated. While RFID is a one-way communication system where data flows from tag to reader, NFC can be used for one- or two-way communication with peer-to-peer capabilities. Two-way communication allows us to share info and pair devices. So far, barcodes and RFID chips have had clear impacts for inventory and tracking applications, and while NFC has found some momentum in similar areas, it has yet to take off with consumers. “It’s an application looking for a market,” says Dean Freeman, a semiconductor expert at Gartner analytics.

ThinFilm is moving towards producing NFC chips using a roll-to-roll process.

If ThinFilm can increase volume, that would decrease the cost of each NFC device, potentially to a few pennies per chip, which is necessary for many potential applications to be successful. “If you’re going to compete against NFC silicon or RFID or silicon or barcodes, the way you do that is with cost, and the way you get cost is with volume,” Freeman says.

A traditional silicon-based semiconductor manufacturing plant (known as a “fab”) produces chips on silicon wafers (usually around 2–12 inches in size). These wafers must be processed in various steps, and the development speed is limited by the number of wafers —“they’re processed essentially one wafer at a time,” says Matthew Bright, the senior director of product & technical marketing at ThinFilm. The equipment in most fabs is extremely expensive, and starting a new facility can cost between $5–10 billion dollars.

In contrast, the new San Jose plant is currently producing chips on square sheets of metal, similar to foil. These stainless steel sheets, at about 12 inches per side, allow ThinFilm to make more integrated circuits at a time as they are able to print on larger areas that are not limited by individual wafer processing. Moving to sheets also means they can start printing directly onto product packages and labels at scale, rather than removing chips from wafers and attaching them. By printing the electronics directly onto labels, companies can ensure that the product is not tampered with, and that counterfeits cannot be given the same tags.

Printed roll-to-roll processes use metal foils instead of standard silicon wafers to produce NFC chips, like the ones shown here.

Presently, traditional wafer fabrication requires a series of up to 15 mask steps that take about one to two days each, resulting in a 30 day lead time for a typical NFC chip. Printed electronics offer the possibility of reducing that time to one to two days total, making it quicker and easier to design and produce new devices.

The next step is to move to printing on rolls of about 220 yards in the next 12-18 months. Moving to a continuous sheet process will theoretically allow ThinFilm to increase production from around 24 million units per year to 5 billion NFC units per year. “They’re actually printing a semiconductor in volume, which no one has really done before,” Freeman says.

For its part, ThinFilm says they do not intend or expect to replace the traditional semiconductor industry, and they do not view their work as competing head-to-head with established players like Intel or Samsung. For now they are relying on a hybrid process, where some steps are completed through printing, and others are done using standard techniques. Bright says their plan is “about complementing existing semiconductor manufacturing capacity so electronics can move into new market segments.”

But some are still skeptical. Dr. Yang expressed concerns about the tradeoffs between cost and quality in the proposed devices. The process of printing chips requires expensive chemical inks and a level of printing precision that has historically been challenging to achieve, resulting in low quality devices.

“Getting a fully printed process has been a challenge for the industry, one they are still trying to overcome,” Freeman says. To date, fully printed devices have not achieved the complexity of traditionally fabricated chips, and less complex options, like barcodes, can cost mere micro-cents. “No one has really been convinced that the electronics [created by printing] are anything to be happy about,” Yang says. Not to mention the fact that the chips used to create the IoT still require power sources—most lower energy ones use batteries which will quickly need to be replaced.

Cheaper Chips

ThinFilm says they are not currently focusing making their chips more complex, but cheaper. By decreasing the cost of NFC production, the company envisions incorporating small amounts of intelligence into billions of products, like the alarms and coffee makers mentioned earlier, or a pill case to confirm if you’ve taken your medication. On the device side, there’s some potential. The number of devices that are NFC enabled has doubled this year, and Apple recently announced that they will expand access to NFC chips in their phones to more third-party developers.

For now, though, the uses haven’t been quite so, well, useful. The cost of ThinFilm’s NFC chips, about 25 cents, are prohibitively expensive for all but luxury goods like designer handbags or expensive alcohol. ThinFilm has run test campaigns with craft brewers like Coronado Brewers—a set of coasters embedded with data about the company—and with the distiller Johnnie Walker—to deliver two different messages depending on whether the bottle is open. They also tagged diamond certificates with NFC chips that, when scanned, informed consumers of the background of the particular gem they were considering. ThinFilm also has been producing anti-theft tags and temperature sensors that can store information on whether the product was within specified temperature ranges during its journey.

As IoT applications expand, NFC chips may become incorporated in household objects like medication bottles.

Beyond simply tagging goods you can buy in a store, IoT industry leaders have their sights set on hospitals, clinics, and other healthcare uses, where the difference between 10 cents and 50 cents is less significant—and might save someone a costly trip to a doctor, say if they are concerned that they accidentally took two doses of their medication and the pill case can confirm they did not. “We believe we’re going to see products in the human monitoring space—like the medical sector—with a decent penetration by 2020,” Marsh says.

But even if manufacturers can make enough of these chips to put them in seemingly everything, including your pill case, IoT presents privacy concerns that have not been addressed. The real appeal of this technology lies in connecting various devices so that the data that has been collected there can work together. This rapidly brings us towards automation—where your blinds could react automatically to changes in sunlight. On some level, this introduces convenience, but it also brings up unsolved challenges that we may not be ready for. Objects that are connected to home networks and collect more and more data about our behavior in our homes and other personal environments raise concerns about data security and privacy. Beyond obvious privacy concerns, allowing companies to collect data about our actions, like monitoring what order we approach aisles in the grocery store, or how long we spend selecting a product, may eventually introduce a level of convenience for the consumer, but also allows the companies to determine behavioral models and increase revenues by finding specific ways to target consumers and optimize store layouts.

We do not yet have systems to effectively and seamlessly integrate all of these devices, and when we do, they will be subject to many of the same privacy and governance issues as our other algorithmically-controlled platforms like Facebook.

For now, these concerns remain more speculative than real since many of the proposed uses for IoT, like tracking goods or recognizing temperature changes, can be done passively in other, cheaper ways. “Who is going to pay 10 cents for an RFID tag for a 20 cent pack of gum? The places where this can go are pretty limited,” Yang says.