Atomic layer deposition (ALD) and atomic layer etch (ALE) are the atomic-scale counterparts to conventional deposition and etch: instead of running a continuous reaction whose rate you try to time, each splits the process into self-terminating half-reactions so the surface changes by exactly one atomic layer per cycle. The result is thickness and depth control at the level of a single monolayer, together with a conformality and uniformity that ordinary flux-driven processes cannot match. As transistors have gone three-dimensional — FinFET, gate-all-around, 3D NAND, high-aspect-ratio DRAM capacitors — these cyclic, self-limiting processes have moved from niche to indispensable, because they are the only way to coat or carve a surface uniformly regardless of its shape.\n\nALD builds a film one saturated monolayer at a time. A cycle exposes the wafer to a precursor pulse that chemisorbs onto reactive surface sites and then stops — once every site is occupied the reaction self-limits, so excess precursor and byproducts are simply purged away. A second pulse of a co-reactant then reacts with that adsorbed layer to form the desired material and regenerate a fresh set of surface sites, and purging again completes the cycle. Because each half-reaction saturates rather than runs to a timed thickness, the film grows by a fixed, material-specific growth-per-cycle, and total thickness is just cycles × growth-per-cycle. The self-limiting nature is also what makes ALD perfectly conformal: deep in a trench or around a fin the reaction still only ever deposits one monolayer, so vertical and horizontal surfaces coat identically.\n\nALE is ALD run in reverse — remove one monolayer per cycle instead of adding one. The first step chemically modifies only the top atomic layer (for example, adsorbing chlorine or forming a fluorinated layer), and because that modification saturates the surface it too is self-limiting. The second step then supplies just enough energy — low-energy ions or a thermal pulse — to desorb only the modified layer, leaving the unmodified bulk beneath untouched. Etch depth becomes cycles × etch-per-cycle, and because the removal energy is kept below the threshold that would sputter the underlying material, ALE causes far less damage, roughness, and selectivity loss than continuous plasma etching. This precision matters most exactly where a few atoms of over-etch would ruin a device: gate recesses, channel release in gate-all-around, and other atomically thin layers.\n\n| | ALD (deposition) | ALE (etch) |\n|---|---|---|\n| Goal | add material | remove material |\n| Cycle | precursor → purge → co-reactant → purge | modify → purge → remove → purge |\n| Self-limiting because | sites saturate with precursor | only top layer is modified |\n| Per cycle | +1 monolayer (growth-per-cycle) | −1 monolayer (etch-per-cycle) |\n| Amount set by | number of cycles, not time | number of cycles, not time |\n| Signature strength | conformality in high-aspect-ratio | low damage, atomic precision |\n| Trade | slow (throughput) | slow (throughput) |\n\n``svg\n\n``\n\nThe price of atomic precision is throughput, and the payoff is 3D scaling. Both processes are slow — they run many pulse-and-purge cycles to build or remove even a few nanometers — so they are reserved for the layers where control, conformality, or damage-freedom actually justify the cost: high-k gate dielectrics and metal gates, diffusion barriers and liners, spacer-defined multi-patterning, DRAM capacitor and 3D-NAND stacks, and channel release in gate-all-around. The tooling is a distinct market: ALD and ALE are dominated by a handful of suppliers (ASM International, Lam Research, Applied Materials, TEL), and process development centers on precursor chemistry, surface saturation windows, and purge efficiency. As dimensions keep shrinking, the fraction of a process flow that uses atomic-layer steps keeps rising, because timed, flux-limited processes simply cannot hit the tolerances that 3D devices demand.\n\nRead ALD and ALE through a control-theory lens rather than a 'slow deposition/etch' lens: the whole point is to convert an analog, rate-×-time process — where thickness or depth is the integral of a reaction rate you can never perfectly know — into a digital, count-the-cycles process where the surface saturates and then refuses to change further. Self-limitation is what removes the dependence on flux, time, and geometry all at once, which is why the same idea, run forward or backward, delivers both the conformal films and the damage-free recesses that three-dimensional transistors are built from. The cost you pay for that determinism is cycle time, so the design question at every layer is whether atomic control is worth the throughput — and as devices go vertical, more and more often it is.