Anaheim, CA. At the International Test Conference, William R. Bottoms, chairman of Third Millennium Test Solutions, delivered a talk Thursday titled “Can We Ensure Reliability in the Era of Heterogeneous Integration?” He focused on quality in an era when devices will include active photonic, electronic, and plasmonic components as well as passive devices, RF sensors, and MEMS.
The focus, he said, is on reliability, not test—test is only a portion of reliability.
He said you may think you have known-good-die (KGD)—but you don’t. It’s simply that the bad transistors have been taken out of loop. And the rate of degradation as function of temperature is rapidly accelerating as we go to smaller geometries and high thermal density. With tens of billions of transistors, it will be necessary to perform continuous test while a device is running.
The fundamental force driving for progress, he said, is the demand for real-time data and processing everywhere all the time. (He mentioned that that could represent either nirvana or a curse.)
IoT and migration to the cloud are changing the global data network, Bottoms continued, adding that the network we have today, built for people-to-people communication, will not continue to work with people-to-machine and machine-to-machine communications—leading to the Internet of Everything.
He cited several examples, including a wetness-sensing diaper that calls your cellphone when it needs to be changed. In the U.S. alone, he said, 40 million diapers per day are used, making this a potentially very high volume, very low cost application.
As for a low-volume high-cost example, he cited a pacemaker with on-board smartphone that can be inserted in the heart via a blood vessel.
Bottoms said the nature of global network traffic is changing as the number of connected devices this year exceeded the number of people on earth. And wireless traffic will surpass wired traffic by 2016. To handle increasing bandwidth requirements, he said, broadband speeds must double in next four years.
He also noted that 70% data traffic stays within the datacenter, where you can achieve low latency and provide thermal management. In contrast, latency is not a particular issue for most IoT nodes, but IoT devices can operate in hostile environments.
Mobile-specific packaging brings reliability challenges, he said, as a thin, small device integrates sensors, cameras, antennas, and so on—possibly using stacked-die technology. “We can build it,” he said, “but can we ensure it’s reliable? Not yet.” The goal will be to provide security and reliability with no increase in cost.
Bottoms went on to explain that networks must change globally and locally, replacing hierarchical architectures, which add latency and power consumption with each layer, with a flat architecture and optical communications from board to package.
Bottoms set as a goal a 104 improvement in performance over 15 years, adding that core packaging technologies like wirebond, flip-chip, and WLCSP will continue to progress. But, he added, new materials will be required. Possibilities include carbon conductors, composite copper (which can offer a 44% improvement in current-carrying capacity), Cu-SWCNT, and low-temperature Cu nano-solder. Such materials exist, he said, but are not yet integrated, and the effort to do that should not be trivialized, although we do have a 15-year timeframe.
3D-SiP packaging with heterogeneous integration will be required, he said, with the integration of separately manufactured components, such as 40-nm, 28-nm, and 20-nm devices as well as MEMS, RF functions, photonics, analog blocks, increasingly complex passive components, and microfluidics. To get the components as close together as possible (to reduce size and latency) it will be necessary to go 3D. And perhaps we will add new types of heterogeneous components that we aren’t even thinking about today.
Next, he said, we will be adding photonics to achieve further reductions in power, latency, and size, but cost and thermal management could be issues. And it will be important to move the photons as close as possible to the transistors. Today, he said, long-haul communications employs the butterfly package for the laser chip, but it’s gold, expensive, large, and has a high parts count; it won’t work for low-cost photonics.
In any event, he said, testing will not be what we use today.
He cited several technologies and products that point the way to the future of heterogeneous integration, including the AMD Fiji architecture, Japan’s Petra project, the IBM Neurosynaptic processor, and the Xilinx Zynq programmable multiprocessor SoC.
Co-integration of technologies, Bottoms continued, should make use of the solution that’s best—and to a large extend that will remain electronics. Plasmonics may have a role to play, he added.
He offered several recommendations:
- Reduce processing steps, such as remove package underfill.
- Reduce power and improve thermal management.
- Continue Moore’s Law scaling.
- Reduce leakage currents.
- Distribute power at high voltage.
- Don’t make heat in first place.
- Improve thermal conductivity with new materials.
- Consider microfluidic cooling, although it may be too expensive.
He concluded by looking at applications of the future, asking whether man-to-machine telepathy is on the horizon. That might seem too much like science fiction, but he cited Emotive Insight brainware as an example of what’s possible today.
“The prospect opens the imagination for science fiction to become reality,” he said, noting that at one time there were no smartphones.
In 15 years, he said, we may have 3D full-color holographic telepresence.
He asked, “We can do it, but can it be cost effective and reliable?” Cost effective-test solutions with intelligent redundancy and continuous test while running will be necessary.
