ZGlue’s secret sauce is its use of a silicon substrate, with a grid of fine wires for reprogrammable connections between the chips, and some resistors, capacitors, and gates for built-in power management, system management, and memory.

To build a ZiP chip, an engineer starts with a cloud-based representation of the grid, and uses drag-and-drop design tools to select from a library of off-the-shelf components. After arranging the various components, the engineer then can use zGlue’s software to specify the various connection paths, and generate a list of the components and the connections along with a software development kit for the product.

But here’s the part that zGlue thinks will really keep the costs of these devices down—the actual routing doesn’t happen until AFTER the components are placed on the ZiP chip. Automated testing tools measure where the chips landed on each ZiP chip’s grid—think of it as a bed of nails—and then make the connections. Built-in circuitry detects which «nails» are connecting to solder bumps on the chip packages. ZGlue’s software takes the map of intended bump positions and the map of actual bump positions and figures out which pin of which package each of the actual bumps is connected to. The router in the software then programs the grid to connect the chips into the desired configuration, programming it into the ZiP chip’s memory. The whole process happens individually for every ZiP chip that comes off the assembly line.

This approach means that the chips don’t have to be precisely placed during manufacturing, so they can be made with relatively cheap equipment and quickly.

“Today, if you want to build a multichip module on a custom substrate, it will take you six months to go from design to getting it through the factory. With our process, we can do it within a week right now, using manufacturing in Taiwan; days if we bring equipment in house,” said Taylor.

ZGlue’s founders—Jawad Nasrullah, Myron Shak, and Ming Zhang—connected with each other at Intel. They trace the ideas that evolved into zGlue’s technology deeper into their semiconductor careers, including work with Transmeta, AMD, and Maxim Integrated. Advisor Taylor contributed to 10 generations of microprocessors during 25 years at Intel, and led a team that developed a hardware random-number generator there.

At Intel, recalled CTO Nasrullah, “We were discussing a future with devices everywhere, but were saying that we would need a salable, heterogeneous way to get there. We learned from what the FPGA guys did in the 80s; maybe FPGAs weren’t the best compared to ASICs, but they were quick and right there.

“Of course, we couldn’t use traditional FPGAs, they are limited to logic, and we needed more. Incorporating off-the-shelf chip-scale packages within the ZiP chip allows designers to put anything in there using their favorite compatible components from their favorite manufacturers. And FPGAs are power hungry, but we were envisioning small battery-powered systems.”

ZGlue officially started in 2014, but things really heated up in late 2015, when the team was admitted to the StartX accelerator, won a development grant from the National Science Foundation, and convinced seed investors to invest. The first prototype ZiP chip came off the line in March 2016; a more fully configured ZiP chip came out this May. ZGlue now has 40 employees and has raised US $10 million in capital to date. While the company is still testing the new chips, it says it’s ready to start volume production. The substrate is being manufactured by TSMC, and the actual modules built on that substrate will be put together by ASE.

“It’s only about 45 days old,” said Taylor, “but we haven’t found problems so far. We would like to do life testing, but have seen no reason to expect to encounter any road blocks.”

Initially, zGlue’s customers will be large companies, and some have already started working on designs, Taylor said. But the really exciting part for those involved in zGlue will be enabling the “two guys in a garage” type of startup to get from idea to market quickly and cheaply—meaning those companies reaching out on Kickstarter won’t need tens of thousands of dollars in preorders before they can get going.

“It will likely cost under $100 to get a prototype from us; of course, it will be much cheaper in volume. But with a prototype, a circuit board, and a 3D-printed case, you can make an integrated device that wouldn’t look funny up against an iWatch. We want to broaden who can build for the IoT,” Taylor said.

With that plan in mind, zGlue demonstrated the technology at a recent Maker Faire in San Mateo, California. It also ran a campaign on Crowd Supply, offering samples to people.  And the team plans to reach out to the small developer community.

One small development team has already gone public with its use of zGlue’s technology: Sentinel Bandage, a project within the University of California at San Francisco’s Surgical Innovations group that is aimed at creating wearables that monitor wound healing.

Nasrullah says while the design tools are already easy for a developer to use, they are going to become even easier—because the tools are based in the cloud, they are generating data about what types of chips developers select, what kinds of buses they use to connect them, and other choices.

“We expect to use that data to train algorithms such that developers can use natural language to describe their intent and have the zGlue software assist in generating the designs,” he said.

And, in the nearer term, pointed out Taylor, the system will be able to give designers feedback, pointing out, say, that they made an unusual choice and identifying the more common selection.