Nanoelectronics strategies from board room to clean room
It was just a few years ago that nanoelectronics was, well, hypothetical. Today it is anything but. The world’s leading electronics companies are pushing the nanoelectronics envelope in attempts to do things like develop faster and denser memory, convert electrons into photons, and forestall, at least for a little while longer, the inevitable demise of Moore’s Law.
Competing against these companies - and, at times, collaborating with them - are a group of agile startups with dreams of being the next Intel. And supplying them with the machines they need are a set of innovative toolmakers seeking to provide the picks and shovels of the next industrial revolution.
They all face difficult challenges, both technological and commercial, regardless of their size or their position in the food chain. Therefore, Small Times invited representatives from the sector to contribute articles on both technical and business hurdles and how they have developed, or are developing, a way over them.
As you will read in the following pages, the problems are many, but the creativity employed in overcoming them is significantly greater.
- David Forman
Nanotech will ride many roads to market
By Tom Theis, IBM
Some experts predict that nanotechnology will be a game changer for the IT industry in general and the microelectronics industry in particular, but major changes in such a large and complex industry will not happen overnight. It looks like incremental improvement of the silicon transistor will continue for ten or more years. Nevertheless, the seeds of massive change have been planted, and I believe that some important innovations in nanotechnology are already on a path to the market.
IBM Research manages a broad portfolio of research projects. At any given time, some of these projects are impacting IBM’s revenue and profits while others are still in an exploratory stage.
As an example of exploratory research with a long-term outlook, researchers at our Zurich Research Laboratory just published their observations of electrical contact formation between a single metal atom and an organic molecule (Science, May 26, 2006, “Imaging Bond Formation Between a Gold Atom and Pentacene on an Insulating Surface”).
An illustration of IBM’s Millipede storage device shows how an atomic force microscope tip inscribes data in a polymer surface. Image courtesy of IBM
Another example is our study of a novel physical process for electrically driven light emission from a carbon nanotube transistor. (Science, Nov. 18, 2005, “Bright Infrared Emission from Electrically Induced Excitons in Carbon Nanotubes”).
We don’t yet know how this fundamental knowledge will be applied, but given the history of miniaturization in electronic devices and the strong market incentives for further miniaturization, we are confident that it will be.
At the same time, many of the projects in our nanotech portfolio are much closer to product applications. MRAM (magnetic random access memory) started out as a highly exploratory project at IBM Research in 1996, but today we are thinking about how to take it to market. Millipede, our nanomechanical approach to mass storage of information, is at a similar stage. And of course, some of our innovations in nanostructured materials and fabrication processes are already in product development. Stay tuned for the announcements.
How will these nanotech innovations go to market? Many will be integrated into existing IBM product lines. IBM is a leading supplier of scientific supercomputers, servers and data storage systems, and a leading designer and manufacturer of microprocessors and other key components of those systems. All of these products and businesses place a high premium on performance and thus demand the continuous introduction of new materials and manufacturing processes, and improved devices for logic, memory and communications. In other words, these products are a prime target for our nanotech innovations.
Not every invention and bright idea from our research laboratories will find its way immediately into an IBM product, so we also license our intellectual property to other innovators. More important, we also partner with others to develop new products and market opportunities. That is why we have pioneered the use of our intellectual property portfolio to help create open innovation networks, where companies can share IP as a foundation for the creation of new products. Take a look at your portfolio of intellectual property, your customers and your partners. Is there something we can each bring to the table that might allow us to create an entirely new product or service?
A successful partnership is likely to be based on a shared vision of potential markets, the ability of each partner to share the cost and risk of collaborative development and the ability of each partner to contribute key resources, talent and expertise. Such partnerships can greatly accelerate the movement from laboratory results toward new products and new markets - that is, they can yield innovation that matters.
Tom Theis is director of physical sciences at IBM’s T.J. Watson Research Center (www.watson.ibm.com) in Yorktown Heights, N.Y.
Bright light from tiny tubes
By Jia Chen, IBM
Today information typically travels as photons in optical fibers deep beneath the ocean and across continents. Yet we access the information as electronic signals through computers, cell phones, Blackberries and iPods. In our wildest imagination, is it possible that one day all information could be ferried by photons, which travel much faster than electrons?
At IBM, we made a novel ultra-bright and ultra-small light source using carbon nanotubes - a breakthrough in nanophotonic devices, and a step closer to transmitting all information with photons. The nano-“flash-light” emits in a wavelength of 1 to 2 micrometers, a range widely used by the telecommunications industry to send information through optical fibers, and 1,000 times brighter and more efficient than previously demonstrated.
Carbon nanotubes are hollow cylindrical tubes with all of their atoms on the surface. They are known to have excellent mechanical and electrical properties. For example, they have 100 times the tensile strength of steel with one-sixth the weight, and can carry 1,000 times more electricity in a tiny area than metals such as silver and copper. Yet this shows that they also have potential optoelectronic applications.
The conventional approach to producing photons from electrical signals is to bring negative charges (electrons) and positive charges (holes) together for them to neutralize each other and emit photons. In the past, electrons and holes were introduced from the two electrodes of a carbon nanotube device separately, and the chances that they would meet each other and emit photons were so low that finding applications for them seemed a distant possibility.
In our new devices, only one type of charge carrier (either electron or hole) is needed to produce light, which is much easier to realize than previous methods.
We played a little trick on the charges (e.g., electrons) moving along nanotubes. We found a way to speed up the electrons by creating a “waterfall” landscape for them, which allowed them to pick up enough energy and create tightly-bound electron-hole pairs. The electrons and holes within the pair will then neutralize each other and emit photons 1,000 times brighter than previously reported. We were able to coerce the electrons to convert the energy to light instead of dissipating into heat.
Our method greatly improved the electron-to-light-conversion efficiency such that every electron injected into the carbon nanotube participates in the light conversion process. A simple way to create the waterfall landscape is to use substrates with different dielectric constants that the nanotube rests on (e.g., by removing part of the underlying support from a nanotube). The new method generates about 100,000 times more photons per unit area per unit of time than large area Light Emitting Diodes, and from an area that is a trillionth of the emitting area of a regular 60W tungsten filament light bulb.
These nanoscale light sources conveniently use the same fabrication processes as semiconductor silicon devices. They have the potential to be built into complex light-based circuitry with the same footprint as silicon electronic components, enabling the integration of both optics and electronics on the same chip. Information can then be ferried not only with electrons, but also with light.
With the aggressive miniaturization of semiconductor chips, the metal wirings currently used to connect the different components on a single chip will suffer increasingly from problems such as lack of speed and unacceptable levels of power dissipation, eventually limiting chip performance. For instance, in a Pentium 4, more than 50 percent of its power is consumed by metal interconnects. These on-chip light sources could eventually provide an attractive alternative as optical connections that generate less heat and support far higher bandwidth than metal wires.
In addition, the light wavelength from the nanophotonic devices can be tuned by using carbon nanotubes with different diameters. Hence one can make nanotube emitters with both infrared and visible light. The devices can also be made on a flexible substrate. The efficient generation of focused light could plausibly be used for carrying out optical probing, providing on-chip variable-wavelength light sources for bio-sensors, and manipulation and spectroscopic analysis at the nanoscale regime where it is impossible to focus light due to physical limitations.
Jia Chen is a research staff member of nanometer scale science and technology at IBM’s T. J. Watson Research Center (www.watson.ibm.com) in Yorktown Heights, N.Y.
The business case for universal memory
By Greg Schmergel, Nantero
Nantero is a product-focused intellectual property company whose goal is to develop NRAM - a high-speed, high-density nonvolatile random access memory. In other words, we want to develop a universal memory capable of replacing SRAM, DRAM and/or flash depending on the application.
In the field of nanotechnology, a product focus is not always the case. Some business plans and even companies get very focused on the technology itself and lose sight of the market, which is easy given how remarkable nanotechnology is. Nantero’s technology is itself quite exciting - we’re using millions and millions of moving nanotubes to store data - but we strive to maintain a focus on the products and our customers. The memory technology used in the end products helps deliver performance improvements and new features. That is what consumers want and expect. We need to deliver memory chips that provide added value over existing and emerging competitors and that become a must-buy for makers of electronic devices.
Electronics designers today choose between several types of memory chips, including DRAM (dynamic random access memory), SRAM (static random access memory) and flash. Each type of memory has its advantages and disadvantages. DRAM is cheap and high-density, but it needs constant refreshing and it’s volatile, so when the power is turned off the data disappears. SRAM is very fast, does not need refreshing, and is easily embeddable in a chip alongside logic, but the cells are very large and it’s also volatile. Flash is non-volatile, but it is comparatively slow, requires block erase, and is not easily embeddable in a chip alongside logic. Almost everyone would prefer to use a universal memory, which combines all the positive attributes of DRAM, SRAM and flash, if only one existed.
Most manufacturers of memory and logic devices are actively seeking to replace the memory they currently produce or embed alongside logic, because they need to eliminate current memory manufacturing and performance limitations. Their end customers are also demanding that the new memory technologies satisfy future end product specifications. This activity requires significant resources and commitment and would be considered successful only if the new memory production cost is competitive and scalable for many years to come.
Given that a universal memory would have a market in the tens of billions of dollars per year, there have been many large efforts to develop one over the past few decades, with two of the most significant being FRAM (ferroelectric random access memory) and MRAM (magnetic random access memory). Neither has made it to market as a scalable and cost-effective memory, so the field remains open.
Nantero’s memory, called NRAM, is intended to combine the non-volatility of flash with the speed of SRAM and the density of DRAM. Importantly, the manufacturing process is simple, with only one additional mask layer, and requires no new capital equipment. In addition, NRAM’s basic concept is scalable down to below 5 nanometers, which means that it would be a viable memory design for decades to come.
We stay in close contact with the end customers and the electronic device manufacturers who would purchase and integrate the memory. There is tremendous excitement across multiple applications about what could be done with a true universal memory and they would all benefit by differentiating themselves from their competitors. Laptops would turn on instantly, cell phones would be more powerful and have a longer battery life, and a variety of new products could be designed, taking advantage of a substantial increase in the amount of memory that could be embedded in microcontrollers and logic chips such as ASICs or FPGAs.
Nanotechnology often carries with it a misperception that it is closer to science fiction than commercial reality, with complex, breakthrough products being decades away. In reality, companies are already in the later stages of developing highly innovative products that deliver benefits that simply could not be achieved without the control of matter at the molecular level.
Greg Schmergel is co-founder and CEO of Nantero Inc. (www.nantero.com) of Waltham, Mass.
Leaping the hurdles to making nanoelectromechanical memory
By Thomas Rueckes, Nantero
Nantero’s goal is to make a memory that combines the non-volatility of flash with the speed of SRAM and the density of DRAM. Our approach involves using carbon nanotubes to make a nanoelectromechanical memory called NRAM. Many industry experts have predicted that carbon nanotubes will play a critical role in the future of the semiconductor industry. However, those experts also tend to predict that this future will not come to pass for a decade or more. This is because of some substantial challenges Nantero has had to overcome before moving carbon nanotubes beyond fabrication of single devices and into mass production.
The first issue has been a major roadblock - carbon nanotubes are grown from a metal catalyst, which is frequently iron, and they are generally grown in dirty environments, leading to the introduction of even more metals. Production semiconductor fabs have very strict requirements and will not allow levels of metal that might contaminate other materials in the fab and damage wafers. So off-the-shelf nanotube material would not be allowed in any CMOS fab in the world. Nantero had to resolve this by developing a process for purifying carbon nanotubes to meet semiconductor industry standards, meaning only a few parts per billion of metal can remain mixed in with the carbon. Having done this, we entered into a partnership with Brewer Science Inc. to enable the supply chain and mass produce this CMOS-compatible nanotube material.
A second major issue is the placement of the carbon nanotubes. A single walled carbon nanotube is approximately 1 nanometer in diameter, and may be a micron in length. Generally, research into carbon nanotube devices is done by growing the nanotubes directly on the wafer and then measuring their properties, but this is by no means a scalable process. And certainly the nanotubes cannot be individually positioned in a mass production process. Nantero resolved this issue by developing a process for spin coating the nanotubes onto the wafers, and then patterning them using lithography and etching. This process is compatible with existing semiconductor process tools and results in nanotubes being located only where required by design.
These issues are examples of how complex it can be to transition from a laboratory environment to a mass production environment. However, once carbon nanotubes can in fact be utilized in production, all sorts of possibilities open up, including Nantero’s NRAM.
NRAM uses carbon nanotube nanoelectromechanical switches to represent bits. In the off-state, the resistance of the bit is in the gigaohm range, whereas in the on-state the resistance of the bit is in the kiloohm range. Thus it is simple to read out the bit and determine its state. The bit state can be changed very rapidly since the carbon nanotubes have a very small mass and move only a very short distance measured in nanometers. And the bits are permanently non-volatile, due to Van der Waals forces which bind them in place in the on-state. Importantly, fabricating NRAM is an elegant process requiring only one additional mask layer, meaning that NRAM is not a costly addition to a process. These are some of the NRAM advantages, which make it a potential universal memory.
In addition, NRAM is intended to be drop-in compatible and integrated easily into existing systems by using existing peripheral circuitry and fitting into existing standard packaging. This means that electronics manufacturers would not have to redesign their laptops, PDAs, cell phones, game consoles, or other devices in any way to take advantage of NRAM.
Thomas Rueckes is chief technology officer of Nantero Inc. (www.nantero.com) of Waltham, Mass.
Making TEM economically viable in semi manufacturing
By Kevin Fahey, FEI
Controlling many of the processes used to manufacture devices will exceed the resolution capability of scanning electron microscopy (SEM) as the semiconductor industry passes through the 65 nm technology node. Transmission and scanning transmission electron microscopy (S/TEM, collectively) provide higher resolution alternatives, but manufacturers have resisted their adoption in mainstream process control and failure analysis applications because of burdensome sample preparation requirements. Perhaps equal in importance to the growing need for better resolution is the need for three-dimensional information, brought on by the increasing complexity of device structures.
DualBeam systems, which combine a focused ion beam (FIB) for cross sectioning, and a SEM for imaging, have been widely accepted for their ability to provide access to the third dimension. Now FIB-based TEM sample preparation promises to be an important enabling technology in the transition from SEM to S/TEM for mainstream semiconductor process control and failure analysis.
As shown in the chart accompanying this article, there are many factors driving the transition to TEM. The chart compares the critical length scales for a number of important processes at the various technology nodes from 65 nm down to 22 nm. Clearly many processes already require TEM for adequate control, even at the 65 nm node. The majority of processes are already in the transition region of the chart and many are in the region where only TEM provides adequate capability.
As semiconductor manufacturing moves to progressively smaller technology nodes, more processes require S/TEM capability to achieve adequate control. Data courtesy of FEI
In few industries is the saying “time is money” more true than for semiconductor manufacturing. Processes are highly integrated and automated. Delays at any point in the process flow translate directly into reduced profitability. The problem is exacerbated by the billion dollar capital investment required for a new fab. Seconds count. Anything that can accelerate the return of an errant process to full yield is valuable. Thus the requirement that manufacturers move from the relatively quick SEM, which can provide results in minutes, to the laborious TEM, which has historically required days, is most unwelcome.
FIB-based sample preparation offers both temporal and economic advantages. A fully integrated tool set can provide first results in less than two hours and produce multiple samples per hour. Although the initial capital investment in a full suite of tools is measured in millions of dollars, cost-of-ownership modeling demonstrates a total cost per analysis as low as $400. Typical cost per analysis for SEM is in the $150 to $200 range.
One of the most important economic benefits of FIB-based sample preparation is the ability to extract a location-specific sample from a wafer and return the wafer to the process. In process development and integration this eliminates variables introduced by looking at different test wafers for each process step. In production it eliminates the cost of scrapping an entire wafer - potentially worth thousands of dollars - in order to obtain a single measurement.
Returning to the equation of time with money, two cycle times are of critical importance in semiconductor manufacturing - the development cycle and the process control cycle. The first determines the time required to bring up a new process or bring a new product into production. Fewer, faster cycles get the product to market first, permitting premium pricing and increased profit.
Once a product enters high volume production, profitability is determined primarily by process yield. Process control seeks to maintain maximum yields. Anything that shortens the process control feedback cycle contributes to profitability by detecting yield excursions sooner, determining the root cause faster, ultimately reducing the length of the excursion and its impact on average yield.
Kevin Fahey is general manager of the NanoElectronics fab division at FEI Co. (www.feico.com) of Hillsboro, Ore.
Enabling the transition to high-res imaging in semi manufacturing
By Todd Henry, FEI
The continuing reduction in device size has pushed the imaging and analytical requirements of many semiconductor processes beyond the capabilities of scanning electron microscopy (SEM), requiring a transition to transmission (TEM) and scanning transmission electron microscopy (STEM, or collectively S/TEM) for mainstream applications. S/TEM provides both atomic scale resolution and much better material contrast but the transition has been impeded by the significantly greater requirements for S/TEM sample preparation - specifically, the requirement for samples thin enough to transmit electrons (100 nm or less).
SEM scans a finely focused beam of electrons over the sample surface and synchronously detects various signals caused by interactions between the beam electrons and the sample atoms. The sequentially acquired signal is assembled into a virtual image that associates signal strength with instantaneous beam location. In semiconductor applications SEM resolution is typically two to three nm. The primary limitation on resolution is beam spreading within the sample.
Focused ion beam-based S/TEM sample preparation uses the ion beam to remove material from either side of the targeted feature until the remaining structure is thin enough to transmit electrons. Images courtesy of FEI
TEM illuminates the entire imaged region of the sample simultaneously with a relatively broad electron beam. It forms a real, magnified image from transmitted electrons using lenses located beyond the sample. STEM may be thought of as a hybrid between SEM and TEM. It scans the sample with a finely focused beam and, using a detector located beyond the sample, constructs a virtual image from transmitted electrons. Both TEM and STEM require very thin samples. The thin sample eliminates much of the beam spreading that degrades SEM imaging and is the primary reason STEM resolution is so much better.
STEM can be performed in some SEMs (at voltages typically less than 30kV) by adding a detector below the thin sample. Dedicated S/TEMs operate at voltages as high as 300kV and can switch easily between modes depending upon specific imaging and analytical requirements. In addition to improved resolution, S/TEM provides greatly enhanced material contrast - essential for distinguishing the many component layers of advanced device designs.
The S/TEM image at right demonstrates higher resolution and better material contrast than the SEM image at left. Images courtesy of FEI
Over the last decade, microscopists have developed new techniques based on the use of focused ion beams (FIB) to expedite sample preparation in order to make S/TEM viable for routine semiconductor applications. These techniques can be highly automated and can offer significant improvements in speed and reliability, as well as reductions in the skills required of the technician.
FIB is similar to SEM in its use of a finely focused beam of charged particles; however ions are much more massive than electrons and can be used to remove (sputter) material from the sample much like a microscopic sand blaster. For S/TEM sample preparation the FIB is used to remove material from both sides of the desired thin section and to cut the section free from the bulk sample.
One of the most important advantages of FIB in this application is its ability to navigate precisely to the location of the targeted feature - often a defect detected by routine inspection. FIB may be combined with SEM in an instrument known as a DualBeam where the SEM is configured to look directly at the FIB cross section, thus providing very fine control of the milling process. This can be critical in determining the proper end-point for milling, especially if the target is a one-of-a-kind defect.
Todd Henry is director of the semiconductor fab business at FEI Co. (www.feico.com) of Hillsboro, Ore.
Survival means learning to adapt
By Barry Weinbaum, NanoOpto
As CEO of NanoOpto, a venture-backed product company, I have learned the hard way during the past five years that you cannot make it without a healthy mix of many ingredients, including good fortune and a little bit of luck. Startups are like teenagers finding their way in the grown-up world. Expectations are often misguided and things don’t turn out the way you initially planned. For success in nanotechnology, as with life in general, evolution and adaptation are critical.
NanoOpto designs, develops and manufactures a broad range of discrete and integrated passive optical components based on our proprietary nanotechnology-based processes. Our products are intended to displace traditional optical components used in consumer electronics and communications products worldwide. Specifically, we serve four markets: digital imaging, for which we make a suite of nano-filters for cell phone camera modules; communications, for which we make a family of optical isolators and discrete polarizers for communications network applications; projection display, for which we make optical waveplates, filters and polarizers embedded in projection TVs and metrology systems; and, optical drives, for which we make a suite of passive optical components for products like DVD players.
To survive, we have had to be extremely flexible in our business model and market approach. We were financed in early 2001 to use nanotechnology to build next-generation optical components for communications networks. In retrospect, we had chosen the absolute worst time to start a telecom components company. Our markets - and our customers - were crashing down around us.
Bold decisions were required, and we used every resource at our disposal to crawl through a narrow porthole and make it to the other side. As a result, we are a vastly different company today. We have adapted our technology and ourselves to receptive markets. We learned through our customers that it does not matter if you are a nanotech company. It only matters that you offer a better and more cost-effective way of doing things, and that you can execute and deliver. Therefore, we chose a market diversification strategy because our technology platform was capable of such a move, and because the economic impact and improved value we could provide mattered to customers.
If adapting your strategy to meet real market needs marks the onset of a company’s “adolescence,” then moving into manufacturing is the teenage years. Over the past 18 months, as we began to generate volume-based revenues, operational issues emerged front and center. We required manufacturing capabilities that could build products in volume, at high yield and at reasonable cost.
It was no longer compelling just to show interesting samples. It became critically important to meet customer commitments with volume product deliverables. Customer relationships would be made or broken with people who invested their own personal capital, believing we would deliver on our claims. For the right market reach, the need emerged for a distribution network in critical geographies.
For us, that meant we had to supplement what we considered to be an already strong team, adding critical new skills to the company and integrating new people into the team. In other words, we had to become a real company - all the while maintaining and growing our investor syndicate to demonstrate increasing value so we would have the runway to achieve our fullest potential.
Five years into the experience, we have moved from an idea to a product development engine, from samples to volume, from concept to quality. We feel the company is poised to become an adult. As a result, the issues we face today have little to do with nanotechnology per se and everything to do with solving basic business problems.
Though it has been over-hyped, nano-technology remains full of enormous potential. Like most technology disruptions, it won’t have a sudden impact. Rather, it will make its mark on everyday society over a long period of time by establishing better or more economical ways of doing things. That’s the path to becoming a successful adult, rather than a young prodigy that ends up as a mere flash-in-the-pan.
Barry Weinbaum is president and chief executive of NanoOpto Corp. (www.nanoopto.com) of Somerset, N.J.