A monolayer of tungsten diselenide is less than a nanometer thick, yet it can absorb light strongly, emit brightly, and host excitons that survive at room temperature. That combination has made two-dimensional semiconductors — especially transition metal dichalcogenides, or TMDs — the most studied material family in excitonic device research.
The problem is that the most impressive TMD results are often obtained on flakes exfoliated from bulk crystals with adhesive tape, a method that produces exquisite quality but cannot scale. A flake a few tens of micrometers across, chosen by eye under a microscope, is not a manufacturing process. To make devices at volume, the field needs wafer-scale growth of uniform monolayers, clean transfer onto target substrates, and device processing that preserves the material’s properties.
This article explains how TMDs are made, why each step is hard, and what progress has been reported toward manufacturing that could support photonics products.
Why TMDs are hard to manufacture
TMDs are layered materials. In bulk form, they consist of stacks of monolayers held together by weak van der Waals forces. Each monolayer is a sandwich: a layer of transition metal atoms such as molybdenum or tungsten between two layers of chalcogen atoms such as sulfur, selenium, or tellurium. The intralayer bonds are strong covalent bonds; the interlayer bonds are weak, which is why bulk crystals can be exfoliated into monolayers.
That structure gives TMDs their attractive properties — direct bandgap in monolayer form, strong excitonic absorption, valley-selective optics — but also their manufacturing challenges. Growth must produce exactly one layer over large areas, with uniform composition, low defect density, and controlled grain structure. A second layer changes the bandgap from direct to indirect and quenches photoluminescence. A vacancy or substitution changes the doping and introduces nonradiative recombination. A grain boundary scatters carriers and excitons.
The requirements are stricter than for many electronic materials because optical properties are sensitive to defects. A transistor can tolerate a certain density of defects if mobility remains acceptable. An emitter or detector that relies on narrow exciton linewidth and high quantum yield is far more sensitive. Manufacturing must therefore achieve not just electrical uniformity but optical uniformity — consistent exciton energy, linewidth, and quantum yield across the wafer.
Exfoliation: the research standard
Mechanical exfoliation — peeling layers from a bulk crystal with tape — remains the gold standard for quality. It produces monolayers with low defect density, narrow photoluminescence linewidth, and high exciton quality, especially when combined with encapsulation in hexagonal boron nitride. Many of the most cited results on exciton physics, valleytronics, and polaritons use exfoliated flakes.
Exfoliation is not a manufacturing method. It is slow, random, and produces flakes of varying size, shape, and thickness. It requires manual identification and transfer, and it does not scale to wafers. Its role is fundamental research and prototyping: it establishes what is possible when material quality is high, and it provides a benchmark against which growth methods are compared.
The field sometimes describes progress as “closing the gap with exfoliated quality.” That gap is real: grown films often have broader linewidths, lower quantum yield, and higher defect densities than exfoliated flakes. Closing it requires improving growth, not just scaling it.
Chemical vapor deposition: the leading growth method
Chemical vapor deposition, or CVD, is the most widely studied scalable growth method for TMDs. In a typical CVD process, precursors containing the metal and chalcogen are vaporized and reacted on a heated substrate, forming a TMD film. Variations include metal-organic CVD, which uses volatile metal-organic precursors and offers better control over composition and uniformity, and powder-based CVD, which is simpler but less uniform.
CVD can grow monolayer films over wafer-sized areas, and it can grow heterostructures by sequential growth of different TMDs. It is compatible with standard semiconductor processing in principle, though the growth temperatures and precursor chemistries must be managed to avoid damaging underlying layers.
The challenges are uniformity, defect density, and grain structure. CVD films are typically polycrystalline, with grain sizes that depend on nucleation density and growth conditions. Small grains mean many grain boundaries, which degrade optical and electronic properties. Defects such as chalcogen vacancies are common, because chalcogens are volatile and can be lost during growth. Achieving monolayer-only growth over a full wafer — without bilayer islands — requires precise control of precursor flux, temperature, and pressure.
Progress has been reported on several fronts: growth on sapphire and other substrates that promote larger grains, use of alkali metal promoters to enhance growth, and metal-organic CVD with improved precursor delivery. Wafer-scale growth of uniform monolayer films has been demonstrated in research settings, but yields, reproducibility, and defect densities remain below the levels required for photonic products. The field is moving from “can we grow a wafer” to “can we grow a wafer with uniform optical quality and high yield.”
Other growth methods: MBE, ALD, and solution routes
Molecular beam epitaxy, or MBE, offers precise control over composition and layer thickness by evaporating elemental sources in ultrahigh vacuum. It can produce high-quality films with sharp interfaces, and it is well suited to fundamental studies and to heterostructures. Its drawbacks are slow growth rate, high cost, and limited throughput, which make it less attractive for high-volume manufacturing than CVD.
Atomic layer deposition, or ALD, and related chemical methods can deposit TMDs at lower temperatures and with good conformality, but the films are often amorphous or nanocrystalline as deposited and require annealing to crystallize, which can introduce defects and limit substrate choices.
Solution-based methods — liquid exfoliation, colloidal synthesis, and inkjet printing of TMD nanosheets — offer low cost and large-area coating, but the resulting films are typically composed of small, randomly oriented flakes with many interfaces, giving poor optical quality compared with CVD or exfoliated material. They may be suitable for applications where uniformity and efficiency matter less than cost and area, such as large-area sensors or coatings, but they are not yet a route to high-performance photonics.
Each method occupies a different point on the quality-versus-scalability trade-off. The field is not converging on a single method but on matching methods to applications, just as it does for materials.
Transfer: moving monolayers without damage
Even when a high-quality monolayer is grown, it is often grown on a substrate that is not the final device substrate — for example, on sapphire or copper foil — and must be transferred. Transfer is one of the most underestimated steps in 2D manufacturing. It must move an atomically thin film over large areas without wrinkling, tearing, contaminating, or introducing strain.
Common transfer methods use a polymer support layer, such as PMMA, that is coated onto the film, after which the growth substrate is etched or the film is delaminated, and the polymer-film stack is placed onto the target substrate and the polymer removed. Variations use thermal release tape, water-assisted delamination, or direct transfer without polymer. Each method has trade-offs: polymer residues can contaminate the surface and degrade optical properties; etching can damage the film; mechanical delamination can introduce strain and wrinkles.
For heterostructures, transfer must also control alignment. Twisted bilayer and van der Waals heterostructure devices rely on precise relative rotation between layers, which determines moiré potentials and interlayer exciton properties. Achieving that alignment over wafer areas, reproducibly, is an unsolved manufacturing problem. Most twisted devices are still assembled flake by flake under a microscope.
Progress on transfer includes cleaner polymers, dry transfer methods that avoid solvents, and automated transfer tools that control pressure, temperature, and alignment. Wafer-scale transfer has been demonstrated in research settings, but uniformity and yield remain challenges. Transfer is often the step where a film that looked excellent on its growth substrate degrades on its device substrate.
Encapsulation and passivation
TMD monolayers are sensitive to their environment. Air, moisture, and substrate interactions can dope the material, broaden exciton linewidth, and quench photoluminescence. Encapsulation in hexagonal boron nitride, or hBN, is the standard method for preserving high optical quality in research devices. hBN provides an atomically flat, chemically inert environment that protects the TMD from contamination and reduces inhomogeneous broadening. Encapsulated devices show dramatically narrower linewidths and higher quantum yield than bare devices.
Encapsulation is not yet a manufacturing process. hBN itself must be grown or exfoliated, transferred, and aligned, adding further transfer steps. Alternative encapsulation methods — atomic layer deposited oxides, polymer coatings, and other dielectrics — are being explored, but they often introduce doping, strain, or nonradiative losses that hBN avoids. The ideal encapsulation would be wafer-scale, low-temperature, and compatible with device processing, while preserving the optical quality that hBN provides.
Passivation of defects is another approach. Chemical treatments that fill chalcogen vacancies or passivate dangling bonds can improve photoluminescence quantum yield, but their stability under device operation and their compatibility with manufacturing remain open questions. A treatment that improves a film for a photoluminescence measurement but degrades under current or heat is not a manufacturing solution.
Metrology: how to know if a wafer is good
Manufacturing requires metrology: fast, non-destructive methods to assess uniformity and quality across a wafer. For TMDs, the key optical metrics are photoluminescence intensity, peak energy, linewidth, and quantum yield, along with Raman spectra that indicate layer number and strain. Electrical metrics include mobility, contact resistance, and on-off ratio.
Mapping these properties across a wafer is time-consuming. Photoluminescence mapping with high spatial resolution can take hours for a full wafer. Raman mapping is similarly slow. The field needs faster, wafer-scale metrology that can be integrated into a production line, along with statistical process control methods that translate optical metrics into yield predictions.
Defect characterization is particularly important. A single vacancy is invisible to most wafer-scale metrology, but a high density of vacancies degrades device performance. Techniques such as scanning transmission electron microscopy can image defects atom by atom but are destructive, slow, and limited to small areas. Correlating wafer-scale optical metrics with atomic-scale defect densities is an active area of research, and it is essential for establishing specifications that manufacturers can meet.
Yields, uniformity, and the path to products
The ultimate test of manufacturing is yield: what fraction of devices on a wafer meet specifications? For photonic devices, specifications include exciton energy, linewidth, quantum yield, and device metrics such as responsivity or threshold. Yield depends on every step — growth, transfer, encapsulation, lithography, contacts, and packaging — and a failure in any step can scrap a device.
Current TMD manufacturing is far from the yields required for consumer electronics. Research demonstrations often report results on selected devices, not wafer-average yields. That is appropriate for physics research but not for product evaluation. The transition to manufacturing will require reporting wafer-scale statistics: distributions of key metrics, not just best-device values, and yield as a function of device type and specifications.
The path forward is incremental. Near-term products are more likely to use TMDs in applications where small areas, relaxed uniformity, or hybrid integration are acceptable — for example, small sensor arrays, research instruments, or niche photonics where a single high-quality device is sufficient. Large-area, high-yield products such as displays or megapixel imagers will require further advances in growth, transfer, and metrology.
For readers evaluating a manufacturing claim, the questions are: What wafer size? What is the distribution of photoluminescence or device metrics across the wafer, not just the best spot? What is the yield for the reported device type? And was the process demonstrated with a method that can be repeated, or with a hand-optimized sample? Answers to those questions separate a growth demonstration from a manufacturing process.
Sources and further reading
This article draws on review literature on TMD growth, transfer, and device manufacturing. Useful sources include reviews in Nature Reviews Materials, Chemical Reviews, Advanced Materials, and 2D Materials on CVD growth of TMDs, wafer-scale synthesis, transfer methods, and encapsulation, as well as textbooks on thin-film growth and semiconductor manufacturing for the broader context of yield and metrology. Specific growth conditions, grain sizes, and defect densities vary widely by method and laboratory, so this explainer avoids citing single-paper values as general manufacturing specifications and describes the trade-offs qualitatively.