The Next Revolution in Semiconductor Processing: Atomic Layer Etching
19 MAY 2022
What will be the next revolution in semiconductor processing? In this blog, Martin reveals what atomic layer processes are, how thermal ALE works, and what we can make using ALE.
What are atomic layer processes?
Since the early 2000s, semiconductor engineers have used highly controlled chemical processes to grow materials with extreme precision. These atomic layer deposition (ALD) processes rely on combinations of chemistries with the special property of self-limitation. In a self-limiting ALD process, one chemical sticks to a surface but does not stick to itself, growing one layer of the material contained in that chemical. Then, a second
chemical changes that surface, so a fresh dose of the first chemical will stick again. By repeating this process, a film is grown to a few nanometers thick. The most exciting part of such a self-limiting process is that complex structures, with lots of pores or pillars, can be evenly coated – an impossible task for other methods of making these tiny layers.
Today, we are on the verge of another atomic-scale revolution in semiconductor design enabled by atomic layer etching (ALE). An ALE process works like an ALD process, except that material is removed, instead of added, one layer of atoms at a time. In ALE, the first
chemical or plasma modifies the surface layer of atoms. Then, a second chemical or plasma turns the modified surface atoms into gas molecules that evaporate from the surface. A good ALE process is highly selective, which means it can etch only one material or one type of material. This is critical for increasingly complex device configurations where multiple materials, including semiconductors, metals, and dielectrics, can be exposed to the etch chemistry.
Many ALE processes use direct plasmas, where the chemistries and plasma process conditions can result in the protection of some materials while other materials are etched. Direct plasma schemes are used, for example, to tune the etching of silicon oxide relative to silicon nitride. Since plasmas are directional, only materials with a line of sight to the plasma source may be etched. However, some processes use a remote plasma, where neutral species that escape the plasma are part of the etch process. ALE methods that rely only on surface chemical reactions – thermal ALE – can be used to thin down materials in almost any geometry. Together, plasma ALE and thermal ALE help engineers shape materials to make complex devices.
How does thermal ALE work?
Different types of materials require different ALE chemistries. Each ALE process involves modifying a very thin layer of the surface, then removing that layer by forming a volatile chemical from the modified surface material.
For ALE of metals and alloys, the surface metal must first be converted into a compound with a higher metal oxidation state. This can be accomplished by oxidation, hydroxylation, or halogenation. Chemicals such as oxygen (O2), water (H2O), or chlorine (Cl2) could be used for this step, but more specialized chemicals can be chosen to meet selectivity or safety requirements. Surface metal atoms with higher oxidation states can readily form complexes with ligands such as beta-diketonates (e.g., hexafluoroacetylacetone, Hhfac) or carboxylic acids (e.g., formic acid), creating volatile metal-organic complexes that leave the
surface and are pumped out of the ALE chamber. A few typical ALE chemical schemes for metals are shown below.
For ALE of compounds where the metal is already oxidized, such as high-k dielectric metal oxides, a different chemical approach must be used. In nearly all cases, a metal oxide surface is first converted into a metal fluoride. For some elements, such as aluminum, zirconium, or hafnium, the fluoride is very stable and will be readily formed by reacting the
oxide with a fluorinator such as anhydrous hydrogen fluoride (HF) or xenon difluoride (XeF2). After the surface fluoride is formed, a different chemical, such as a chlorinating agent (e.g., dimethylaluminum chloride, DMAC), is introduced. This agent exchanges a ligand (such as chlorine) with the fluorine at the surface, forming a volatile metal chloride which leaves the surface along with the modified ligand exchange agent. Sometimes, to help
control the reaction, the surface layer is first converted into a different metal oxide material before the fluorination step. A few typical ALE chemical schemes for metal oxides are shown below.
What can we make using ALE?
As transistor designs evolve to reach few-nanometer technology nodes, ALE will be needed to both sculpt and connect transistors. Just one or a few ALE cycles could be used to “clean up” the surfaces of transistor materials after they are deposited or processed, removing tiny artifacts that could cause unwanted performance problems.
ALE could also be used to help pattern the tiny wires that connect these transistors. The formation of these wires, made up of lines and vias, requires many patterning and lithography steps, in which some variation in the position of each pattern is inevitable. However, if the smallest patterns are displaced by even a few atoms, then some metal vias could be connected too close to the wrong lines, and the reliability of the whole chip could suffer. ALE can be used to remove tiny amounts of metal from the lines before via patterning steps, forming an indentation that helps guide the via to the correct line. For more details on this fully self-aligned via (FSAV) scheme, please see here.
To make the most of growing processor power, ALE can help shape advanced memory devices. As dynamic random-access memory (DRAM) devices become more densely packed, the high-k dielectric film in DRAM capacitors must become thinner. However, it may not be possible to simply grow and crystallize extremely thin high-k films to achieve the properties that provide good dielectric performance. ALE offers a solution: a thicker high-k film can be grown and crystallized, and then some of that film can be removed by ALE to reach the required thickness for the device. This scheme is shown below.