Vật liệu mới đang có nhu cầu cao
Materials suppliers are responding to the intense pressures to improve power, performance, scaling, and cost issues, which follows a long timeline from synthesis to development and high volume manufacturing in fabs. The advances in machine learning help present a wide field of candidates, which engineers then narrow to potential use.
When building standard logic semiconductor chips, the primary materials are obvious — silicon, silicon nitride and oxide and metals. They go into packages made from resin and metal bits. But that brief synopsis sharply understates the number of other supporting materials necessary to build a finished device. Even though the chip industry has been at this for years, there is seemingly no end to the number of new materials necessary for advancing wafer and packaging processes.
Some of those materials are temporary and will be removed during processing. Others provide alternatives to silicon nitride or silicon dioxide to achieve better etch selectivity. Still others act as hard masks, or as adhesives, or any of a number of other prosaic roles. All are critical, but depending on the goals of a new material, they may take years to develop. Looking ahead is essential for ensuring that new silicon- or packaging-process development proceeds smoothly toward production.
The need for a new material may become obvious when it underlies a new process or process variant, but there are many other reasons why different materials are required. “We worked with a customer that needed to qualify a new resist because the original resist was no longer available,” said David Park, vice president of marketing at Tignis. Regardless of the motivation, new-material development can be a lengthy process.
Fig. 1: Wafer thinned for use in 3D stacking for advanced packages. Source: Brewer Science
Companies such as Brewer Science exist primarily as purveyors of materials that are necessary to the semiconductor industry. Fabs and semiconductor equipment makers rely on these companies to enable new processes that require a different set of inputs. Although the different materials may serve very different purposes, their development features much commonality.
A broad definition of materials
Ask different materials vendors what constitutes a new material, and you can get a variety of answers. The most obvious one would be the synthesis of materials that never existed before. That’s actually relatively uncommon. More common is the finding of new ways to create existing substances using a more effective or efficient process.
Some of the materials being developed are tweaks of existing ones, adjusted to meet new requirements. Others may represent a new combination of elements that isn’t a new molecule. The reach of consideration has expanded greatly. “Twenty years ago, we were looking at 7 to 10 elements in the periodic table being used in a chip,” said Anand Nambiar, chief commercial officer at EMD Electronics, the Electronics business sector of Merck KGaA Darmstadt, Germany in the U.S. and Canada. “Today there are at least 70 or 75 in high-volume manufacturing, or in some form of development.”
Historically, most of these materials — precursors, photoresists, masks — have been organic. But metal oxides are starting to play a bigger role in patterning and etching. “Traditionally, Brewer has designed organic materials and added silicon materials over the past 20 years for hard mask applications,” explained James Lamb, corporate fellow at Brewer Science. “But we have added metal hard masks, as well, so we’re going into inorganic areas now.”
Because many of the materials already exist, patenting the new substances themselves isn’t typical. What’s important is the means of making them, and process patents are more common. But that recipe, the material, and how it’s applied form a unit that must be considered together. In some situations, the same molecules may be used in different places or layers, but with different thicknesses. Taking the recipe into account, some consider those to be different materials, as well.
Important material properties start with their performance in their application, but they don’t stop there. Acquisition, deposition, side reactions, and many other considerations affect the suitability of a substance for use in high-volume manufacturing.
Hot materials for hot new processes
Conversations with materials companies reveal the kinds of materials being researched today. Brewer Science, for example, has been improving packaging underfills — materials that provide physical stability when building a packaged chip. They must be electrically inert. “Our goal is to make materials that don’t have to short anything,” said Rama Puligadda, CTO of Brewer Science. “They can’t be electrically conductive. They have to be thermally conductive.”
Spin-on carbon materials enable improved patterning when using thin photoresist. A critical requirement here is they must be stable at high temperatures, such as those experienced during depositions. “These things can’t be outgassing, as it will pop and crack the chemical vapor deposition (CVD) material or the atomic-layer deposition (ALD) materials you’re putting down at high temperatures,” explained Lamb.
New battery chemistries require new materials almost by definition. “The range for a vehicle or fast charging in one way, shape, or form comes down to materials,” said Puneet Sinha, senior director, global head of battery at Siemens Digital Industries Software.
Atomera uses a rather common material — oxygen. The novelty lies in how it’s applied and used. During epitaxial silicon growth, for instance, the wafer is exposed to a puff of oxygen. There’s not enough oxygen to create SiO2, which requires two oxygen atoms per silicon atom. Instead, silicon atoms at the surface have dangling bonds that can attach to a single oxygen atom.
“We take oxygen atoms and we dope them on a single layer,” said Scott Bibaud, CEO of Atomera. “But it’s not an oxide. It’s a partial monolayer of oxygen. So you have a perfect silicon lattice down below, and then you have this distorted field in the middle where we’ve got our oxygen atoms. But then, above that, you can continue to grow perfect silicon.”
This stack has several uses. Because the bonds have some mechanical freedom to rotate, it can act as a transition layer between two materials with different coefficients of thermal expansion (CTEs), helping to prevent cracking or other reliability issues. Gallium nitride (GaN), for example, is grown on silicon and has a tendency to crack when cooling. Atomera says its oxygen layer can relieve the stress caused by the mismatch in stress and CTE between GaN and silicon. In a transistor gate, some of the oxygen can float up to clean up the gate oxide transition. It also can act as an impurity getter. And it can reduce transistor variability.
Another key material development involves the fabrication of EUV pellicle (the material that protects extreme UV masks) and lens coating. Canatu (named after Carbon nanotubes, or CNTs) creates CNT-based films. Some are employed as pellicles for extreme UV (EUV) lithography. Others are used as lens heaters, forming a film over a camera or lidar lens. The company also is evaluating their usefulness in sensitive sensors given the large material surface area.
In EUV pellicles, the process is the main development. CNTs are not new, but building them at commercial scale is difficult. Canatu says it has developed a floating-catalyst CVD process that can create its CNTs in two steps. The company says this compares with 9 steps for a competitor, which creates a large amount of CNT powder that must then be filtered. The conventional metal silicides used in EUV pellicles today require more than 100 steps.
“We have two different types of reactor,” said Juha Kokkonen, CEO of Canatu. “We have optimized one reactor for the semiconductor domain, where the most important elements are a clean process and a uniform freestanding network. Durability (since pellicles experience high-g forces when the scanner moves), high UV-light transmissivity, and tolerance of temperatures up to 1500°C are required. Lens heaters require a different set of features. “For camera and lidar sensors, we are optimizing electrical capabilities like high conductivity and slightly bigger volumes.”
Entegris makes a material that addresses the challenge of establishing the work functions of nanosheet gate-all-around transistors. “At the transistor level, you adjusted the work function [in the past] by adjusting the thickness of the metal on top of the high‑k dielectric, and that thickness went up to 150Å,” explained Paul Besser, senior director, advanced technology engagements at Entegris. “But between the nanosheets, the spacing is only 100Å.” Instead, Entegris makes dipole shifters — dopants that alter the energy-band structure — to perform this function.
Another material Entegris is working on is an etch stop for use when building power delivery on the backside of the wafer. Wafers must be thinned from 700mm down to 50mm and below, but chemical-mechanical polishing (CMP) would take too long. Grinding is performed for the bulk of the removal, cleaning up with CMP and plasma/wet etching at the end. To prevent etching all the way through the wafer, a clear etch stop is necessary. Entegris is working on a SiGe layer within the wafer that would provide that etch stop.
Entegris’ process challenge is to employ a solid precursor instead of the more common liquid ones. The solid is sublimated — moving directly from solid to gas — and the gas must make its way to the wafer without any side reactions along the way. “If you volatize at a certain temperature — say, 150 degrees — if any place along the path to the wafer is lower than 150 degrees, it will deposit,” said Besser.
Taking the shortest possible route
Creating a new material — or a new application or process for an existing material — comes with several options. Tweaking an established material is the fastest path, assuming no unexpected issues arise. If that’s not possible, then an entirely new synthesis pathway must be discovered, possibly (but not usually) leading to a new material. “If [the requested properties] lie within the range of modification, then we can do that,” explained Hidenori Abe, executive director, electronics business at Resonac. “But if they require two, three, or five times performance improvement, then we’ll need to design from scratch.”
The chosen path may depend on the how well a material or process is entrenched. “In a deposition precursor, you’re constantly trying to find newer molecules,” said Nambiar. “Compare that to a photoresist, where it’s a formulated material. You’re not changing the major components, but you’re trying to tweak the little additives to make it slightly better.”
“We will almost always start with what we know,” said Brewer Science’s Lamb. “That’s the quickest transfer. But in our process, we don’t usually just do that. We look for alternative platforms/chemistries, as well, unless somebody wants thicker or thinner versions of an existing material.“
More typical would be adjusting specific material properties such as heat tolerance or viscosity. Gases are a little different in that they are individual molecules, and assuming those molecules already exist, development work deals largely with the logistical challenges of delivering a uniform, predictable quantity of gas to the desired surface.
Customers drive development, but they don’t always know exactly what they want. “Sometimes they come to us and say, ‘We want this material with this ligand on it. Can you make it?’ Alternatively, they have a specification,” said Abe. The properties in that spec form the requirements for a new material without specifying the material outright.
Partnerships are necessary
Development is typically a collaborative effort. Materials don’t exist in isolation. Instead, they interact with the equipment used to apply them. A new process typically involves equipment changes and new materials. Chipmakers traditionally worked separately with the equipment companies and materials suppliers, but they find that integrating the material with the equipment late in the process often ended up necessitating material rework.
“We used to bring our own materials in, and then it sometimes didn’t yield,” explained Entegris’ Besser. “We had to go back to the equipment supplier and say, ‘Can you make this yield?’ And that led to a time where you didn’t have equipment that works.”
What’s more common now is that the fab customer will work with the equipment vendor directly, and the equipment maker will in turn work with the materials company so that the equipment/material combination works well. “So Lam or Applied Materials would develop the material with us, and they both introduce it to the customer,” said Nambiar. The customer then picks between the equipment/material options.
Materials development is not a fast process. “Most new materials take years of development, perhaps years to even introduce it,” said Nambiar. Canatu’s EUV-pellicle development took seven years. “We listen to what [customers] want to do maybe five years or 10 years later, and then we work together on the proof of concept,” said Abe.
Any new project must include thorough initial research to understand what patents may already be in place. Although materials companies may wish to be self-sufficient, they may run into requests for materials that are already covered by patents. A company can do the development on its own if it can find a synthesis recipe that skirts around an existing patent. If that’s not possible or practical, then it may need to partner with the other company to license the process or to obtain precursors from it.
Design by experiments
The materials-development process involves a combination of experimentation and simulation. Most materials aren’t directly designed but rather emerge from measurements and data. “I would still say [materials development] is more empirical than not,” mused Lamb.
Atomera, for example, developed its technology using ab initio simulation, working from first principles. Other projects rely on empirical data and well-designed experiments to zero in on the best recipes. Merck/EMD can divide up a wafer and place different film stacks on different parts of the wafer to reduce cost and speed up the cycles of learning.
Machine learning (ML) helps to construct and run experiments. “We use a lot of different models to feed our ML algorithm that’s running the DoE [design of experiments],” said Nambiar. “That ML engine gives us hundreds and thousands of possibilities, of which several are real possibilities. And when a chemist looks at it, they can say that those are nonsense and these are real ones. And it would have taken us 100 years to come up with those kinds of options.”
Automation helps as well. “With continuous flow, you can run 100 different reactions in an hour or two, and with inline analysis it’ll feed all that back in the system,” said Lamb. One can automatically use the results to direct further exploration.
Data mining can be particularly valuable for companies with a long history of development. Capturing all that data — even from projects that didn’t pan out — can be valuable when looking into new requirements. Some of those discontinued projects could see a new life. “We’ve really modernized our manufacturing and our pilot line to allow better recording of what happens when we run reactions,” said Lamb. The company has a data-mining tool “… where we can take all our generated information from all our batch works, testing, and evaluations and mine that for directions out of our current material sets.”
Some companies see keep-out zones
Not all companies are comfortable working with all substances. Two considerations apply in particular: safety and sustainability.
Building semiconductors has always involved some dangerous chemicals. College classes decades ago warned of the dangers of, for example, the hydrofluoric acid used as an etchant in the days when human hands moved wafers and cassettes between stations.
Some materials may have changed since then, but there are still risks. “The industry is introducing metal organic resist, and it’s a tin-based molecule,” said EMD’s Nambiar, pointing to another example. “Depending on what type of tin oxide you’re talking about, the toxicity can be different.” Although companies will typically try to focus as much as possible on non-toxic material inputs and outputs, that’s not always possible.
Pyrophoric substances, such as silane, are another example of a dangerous material — not because they’re toxic in their own right, which they may be, but because when placed in contact with air or water they can spontaneously combust. That creates a whole set of logistical challenges, because safety considerations must involve both the material in place and as it’s being deposited, as well as throughout transportation and storage.
As a result, some companies try to avoid the logistics of some substances having these dangerous characteristics. “We don’t handle pyrophorics,” said Besser. He explained, for example, that when using aluminum, “… you could do it from TMA [trimethyl aluminum], which is a pyrophoric, and we just don’t want to handle those because of the health risks.”
Sustainability also involves environmental, geopolitical, and human rights considerations. The extraction of some raw materials can run into any or all those challenges. Lithium and cobalt, for example, have known issues surrounding the working conditions of miners. China controls many of the rare-earth sources, making them politically riskier.
“We have no conflict minerals at all,” said Besser. “We avoid geopolitical regions. All our suppliers sign a business agreement that they will not employ any conflict minerals when supplying us. We have turned down opportunities for business for the safety of our employees and our supply chain.”
Other materials are being phased out owing to environmental concerns. “We also use PFAS [per- and polyfluoroalkyl substances] in certain materials,” said Nambiar. “And our customers are asking us to remove those PFAS.”
Scaling up can’t be taken for granted
Once a new material or synthesis pathway is ready for production, some means of moving from lab quantities to commercial quantities is necessary. “It’s one thing to do research, and another to scale it up and deliver it safely at the right cost,” said Nambiar.
Others agree. “Scaling from flask to production is one of the biggest risk factors you face,” noted Lamb. “You can do most everything in the lab with not too much cost. But once you start scaling the pilot lines, we have to build a new pipeline and manufacturing capabilities to support it.”
That scale-up process might not be trivial. The sorts of reactors and other infrastructure producing small quantities may not work as well in larger batches. For example, heating a volume of gas to a uniform temperature may be harder in a large reactor with external heating elements than for a small one, because the center of the reactor may have a harder time staying at temp. New analytical methodologies also may be necessary to ensure the quality of a material once it is produced in high volumes.
It might not be possible to plan for scale-up at the outset of a project because the necessary process and equipment might not be known yet. But as experimentation starts to home in on a solution, full production must be considered as soon as possible, because it may require new equipment or a relationship with another vendor. Lab equipment and production equipment are often produced by different companies, and new equipment must be thoroughly vetted. “We have to cross-check the construction materials because we always have to take care of ionic contamination,” said Lamb.
Raw-material availability also could be a consideration. “Those raw materials may or may not be anything close to semiconductor grade,” noted Lamb. “If you need a property they give you, they may need intense purification. And that takes a lot of time and adds cost.”
The need to scale up indicates a success in developing a new material. Some projects may never make it that far due to a move in the market rather than a failure of the material itself. Before EUV lithography became production-worthy, there were efforts to develop an interim lithography wavelength of 157nm. That technology gave way to the eventual readiness of EUV, and 157nm never made it to production. Even the development of EUV, which was ultimately successful, involved a number of alternative approaches, most of which went nowhere. The constant changes were challenging for those developing materials.
Humans in the loop
Ultimately, material development is a process that relies on humans for guidance, for knowledge, and for expertise. It’s an area replete with PhDs working on arcane concepts, but simply having a PhD isn’t enough. “We try to hire a lot of PhD chemists and make sure they’re not all, say, lithography [experts],” said Lamb. “So we have inorganic chemists, we have metal organic chemists, we have a pure synthetic chemist, and material scientists. You get a lot of cross pollination of ideas, and that really helps.”
It’s also a process that benefits from so-called tribal knowledge. Those vaults of experimental data are essential, but the experience of the team mining them is just as important. Years of materials development create instincts and intuitions about potentially useful directions that a more naïve approach can miss.
Given the ongoing challenges of a post-Dennard-scaling world, clever ideas are increasingly necessary to maintain progress in semiconductor costs, performance, and power. Because many of the ideas explore uncharted waters, it’s a good bet that new materials will be necessary to move them into production. The need for companies and engineers with materials skills should remain unabated for a good long time.
Related Reading
The Challenges Of Upgrading Lithium Batteries
Safety and energy density are prime motivators as researchers seek to improve lithium batteries.
Key Technologies To Extend EUV To 14 Angstroms
Alongside high-NA EUV will be better-performing photoresists, reduced roughness using passivation and etch, and lateral etching to reduce tip-to-tip dimensions.
Precision Patterning Options Emerge For Advanced Packaging
Photolithography is still mainstream, but innovative new solutions are coming.
Single Vs. Multi-Patterning Advancements For EUV
EUV patterning has come a long way in the past five years, but old challenges resurface with high-NA EUV.
Controlling Warpage In Advanced Packages
Mechanical stresses increase with larger sizes and heterogeneous materials.
