Atomic layer deposition and chemical vapour deposition are not interchangeable. This guide explains how each grows a film, what conformality and throughput mean for your research, and how to choose when the budget funds only one tool.
A department gets one line in the development budget for a deposition tool. The proposal committee asks whether the specification should say ALD or PECVD, and in most drafts the two are written as though a purchasing officer could pick either. They grow films by related chemistry, they are often sold by the same manufacturer, and they sit in the same corner of the cleanroom. What they let you do is different enough that choosing the wrong one costs a research group three or four years.
This article sets out how each process works, what conformality and thickness control actually buy you, where throughput bites, and how to decide when the money funds one system rather than two. It is written for the person who has to defend a specification in front of a procurement committee, not for someone who already runs a fab.
In chemical vapour deposition, reactive gases flow into a heated chamber together, react at or near the substrate, and build a film continuously for as long as the gases keep flowing. Growth rate is set by temperature, pressure, gas flow and the local supply of reactant. In plasma enhanced CVD, an RF plasma breaks the precursor molecules apart, so the reaction proceeds at a much lower substrate temperature. That matters when the wafer already carries metal, polymer or a partly finished device that cannot see high heat.
Atomic layer deposition takes similar chemistry and splits it into two half reactions that never meet in the gas phase. Precursor A is pulsed in and binds to the surface until every available site is taken, then stops: the reaction is self-limiting. The chamber is purged. Precursor B is pulsed in, reacts with what is bound to the surface, and stops. Purge again. That is one cycle, and it lays down a fixed fraction of a monolayer. Film thickness is therefore not a matter of timing a growth run. It is a matter of counting cycles.
That single difference, self-limiting surface chemistry against continuous flux-driven growth, generates every other difference in this article.
Because each ALD half reaction stops once the surface is saturated, film thickness ends up the same everywhere the precursor gas can reach. Trench sidewalls, the underside of an undercut, the inside of a porous membrane, individual fibres in a textile, loose powder tumbled in a drum: all coated to the same thickness. Structures that CVD closes off at the mouth are coated to the bottom by ALD, provided the pulse and purge times are long enough for gas to diffuse in and out.
CVD growth depends on the local arrival rate of reactant, so a deep trench receives more material at the top than at the bottom. Push the process and the trench necks over and traps a void. Across the wafer, uniformity is a function of showerhead design, flow pattern and temperature gradients, and it is exactly what a well built tool and a properly developed recipe buy you.
| Attribute | ALD | Thermal or plasma CVD |
|---|---|---|
| Growth mechanism | Self-limiting surface reactions, one cycle at a time | Continuous reaction driven by gas supply |
| Thickness set by | Number of cycles | Time, temperature, flow and pressure |
| Practical thickness range | Roughly 1 nm to a few tens of nm | Tens of nm up to several microns |
| Conformality in deep features | Coats whatever the gas can reach | Falls off with depth, risk of necking and voids |
| Run to run repeatability | High, because cycles are counted | Good, but tied to chamber condition |
| Typical films | Al2O3, HfO2, TiO2, ZnO, TiN | SiO2, Si3N4, oxynitride, amorphous silicon, DLC |
| Substrate temperature | Low to moderate, plasma variants lower still | High for thermal, much lower for plasma enhanced |
The repeatability line deserves attention in a shared university tool. A film specified as three hundred cycles comes out the same thickness next month, next semester, after a chamber clean, in the hands of a different student. A CVD film specified as ninety seconds will not, unless somebody maintains the chamber and re-qualifies the recipe. In a lab where the operator changes every year, that is a genuine argument for ALD.
ALD is slow by design. Each cycle deposits a fraction of an atomic layer and includes two purges, so a film of a few tens of nanometres runs to hundreds of cycles. Ask for a few hundred nanometres and the run stops being practical. PECVD, running continuously, puts down that same few hundred nanometres in one short run and can reach microns without complaint.
The honest framing is therefore not which technique is better. It is which thickness regime the group's research lives in. Interfaces, gate dielectrics, barrier layers and surface passivation a nanometre at a time: ALD. Structural films, hard masks, waveguide cores, encapsulation and anti-reflection coatings: CVD.
Consumables differ too, and this is where Bangladesh specifics begin. ALD precursors arrive in small bottles, carry a stated shelf life, and several are restricted for air transport. Trimethylaluminium, the standard route to alumina, is pyrophoric. That means an import chain, a temperature controlled store and a plan for what happens when a bottle empties in the middle of a PhD. CVD process gases arrive as cylinders from a gas supplier, which is a simpler chain, except that silane is pyrophoric and drives the entire design of your gas cabinet, detection and abatement. Neither choice is free.
Look at the research the department actually intends to do over the next five years and the answer usually falls out on its own.
Work through the following in order. They are ranked by how often each one settles the question in practice.
Oxford Instruments builds both sides of this choice, which becomes convenient when the two systems eventually sit side by side: one service relationship, similar control software, and a chamber platform your technician already understands. Vvon supplies Oxford Instruments sputter and PECVD systems and the PlasmaPro etch platform in Bangladesh, so a phased plan can be written against real lead times instead of a guess.
Both need process cooling water at a controlled temperature, exhaust to an abatement system, clean dry air or nitrogen for pneumatics and purges, stable power, and somewhere to put the roughing pumps that is not the process room. Get the utility matrix from the manufacturer before the civil works package goes out, not after the slab is poured.
The specification that survives a tender challenge describes the films you need to grow, not the machine you already picked.: Vvon engineering team, on nano-fabrication procurement
If the department expects a second grant cycle, do not spend the first one on a single tool with no supporting infrastructure. Spend it on one deposition system plus the gas, exhaust, cooling and power scope that the second tool will also use. The marginal cost of running an extra gas line and an extra cooling drop during the first fit-out is small. Breaking into a finished cleanroom two years later is not, and it takes the room out of service while you do it.
The corollary is that the first purchase should be the one with the harder site requirements, because that is the one that sets the building scope. In most Bangladeshi departments that is the PECVD system, which is usually also the one that serves the larger number of users.