Poly Process

The poly process is low-pressure chemical vapor deposition of polycrystalline silicon — a film of silicon grains, not a single crystal, that became the default gate electrode material for roughly four decades of MOS manufacturing. It is deposited from silane gas, doped to make it conductive, and patterned into the single feature on a chip that defines a transistor's switching length:

Gate oxide grown
Thermal SiO2, nm-scale
LPCVD poly deposition
SiH4 pyrolysis, 580–650°C
Make it conductivein-situ doping or implant, pick one
In-situ phosphorus doping
PH3 added during deposition
Undoped poly + implant
POCl3 diffusion or ion implant
Litho + gate etch
Defines channel length
Anneal / activate
RTA, grain growth

Polysilicon won the gate-electrode job from metal for one specific electrical reason. Early MOS transistors used aluminum gates, but aluminum's low melting point (660°C) meant the gate had to go down *after* the source/drain dopants were already driven in by a high-temperature anneal — and aligning a metal gate precisely over a channel defined earlier is extremely hard. Polysilicon melts near 1400°C, so it can be deposited and patterned first, then used as the implant mask for the source and drain dopants themselves. The gate automatically ends up perfectly self-aligned to the channel it defines, with zero extra alignment step — this "self-aligned gate" process is the single innovation that made polysilicon gates the industry standard starting in the late 1960s and held that position until high-k/metal-gate stacks took over around the 45 nm node for different reasons (gate leakage and depletion, not alignment).

The film is made of grains, and grain structure is not a cosmetic detail. Polysilicon deposited below about 600°C comes out amorphous and then crystallizes into small grains during the anneal; deposited above roughly 600°C it crystallizes directly as a columnar, grainy structure. Grain boundaries scatter dopant atoms and current carriers, which raises resistivity compared to single-crystal silicon of the same doping level — this is why heavily doped poly (often further strapped with a silicide layer like WSi2 or TiSi2 on top) is needed to keep gate and interconnect resistance acceptable, and why grain size control during deposition and anneal became its own process discipline.

Self-Aligned Polysilicon Gate the gate is deposited first and doubles as its own implant mask ① Gate patterned, before implant Silicon substrate — no junctions yet gate oxide poly gate dopant implant — blocked directly under the gate ② Result: perfectly aligned junctions source drain channel poly gate junction edges line up exactly with gate edges because the gate withstands the implant/anneal temperature, no separate alignment step is needed — the gate IS the mask aluminum gates couldn't survive this sequence, which is exactly why polysilicon displaced them

Poly isn't only a gate material — it's also the fab's own resistor and local-interconnect layer. A second polysilicon deposition (poly-2, common in older BiCMOS, DRAM, and flash flows) stacked above the first became the standard way to build on-chip resistors, capacitor plates, and fuses, simply by controlling doping concentration to dial in a target sheet resistance. Flash memory took this furthest: the floating-gate cell stores charge on an electrically isolated poly layer sandwiched between oxide, using polysilicon's conductivity and compatibility with standard CMOS processing to build nonvolatile memory without any exotic materials.

High-k/metal-gate retired poly from the critical gate role, but didn't retire the process. Below roughly the 45 nm node, polysilicon gates on SiO2 started leaking too much current through the thin oxide and suffered "poly depletion" — a layer of the poly near the oxide interface that doesn't fully invert, effectively adding unwanted capacitance in series with the gate. Replacing the gate stack with a high-k dielectric and a real metal electrode (first gate-first, then the now-standard gate-last/replacement-metal-gate flow) fixed both problems — but polysilicon dummy gates are still deposited and later etched away in that same replacement-gate flow, and poly deposition remains essential everywhere else: resistors, local interconnect, DRAM word lines, and flash.

Patterning the poly gate is its own high-stakes etch, because the film sitting directly underneath it is the gate oxide the whole transistor depends on. Seeing the real etch filmstack explains why this step can't be timed blindly the way a field-oxide etch sometimes can:

Poly Gate Etch Filmstack a few nanometers of gate oxide is all that separates a clean etch from a ruined transistor Cross-section, top to bottom BARC / hard mask cap Polysilicon gate 400-2000 Å etches via Cl2/HBr chemistry Gate oxide — 1-3 nm Silicon substrate a few seconds of over-etch here punches through to Si — this is the whole problem the etch has essentially zero margin once it reaches the gate oxide — hence a dedicated soft-landing step, not a single fast etch

That zero-margin interface is why poly gate etch runs as a deliberate three-stage recipe, switching recipes in response to a live signal rather than a clock:

Poly Gate Etch — Main Etch / Soft Landing / Over-Etch power and chemistry both step down as the etch approaches the oxide Main etch fast, high power Cl2/HBr Monitor SiF* / CO* poly byproduct signal falls Approaching oxide? bulk poly nearly cleared Soft landing + O.E. still bulk poly on the flats — keep running main etch Why step the power down instead of just stopping sooner CD and sidewall profile are still being set as the last nanometers of poly clear a gentler soft-landing step keeps that control intact while giving up almost none of the oxide's margin main etch sets throughput; soft landing and over-etch protect the single interface the whole device depends on

And isolated from the recipe-stage diagram, here is the signal a poly etch tool actually watches to decide when to make that switch:

Poly Gate Etch Endpoint Trace silicon-fluoride byproduct emission vs. etch time — the signal that triggers the soft-landing switch etch time (s) SiF* emission intensity switch to soft landing bulk poly etching, steady byproduct signal low-power over-etch — oxide protected the signal drops well before the oxide is reached, giving the controller a safety margin to step power down in time

That margin is also why poly etch recipes are tuned around selectivity, not just speed. A main-etch chemistry is chosen to clear poly quickly, but the soft-landing and over-etch steps are chosen specifically for high poly-to-oxide etch selectivity — often 100:1 or better — so that even a few extra seconds of plasma exposure after the signal fires removes almost no oxide at all. That selectivity number, not the etch rate, is the real engineering target once the recipe reaches this final stage.

Read the poly process as the step that solved an alignment problem so completely it became structural: by making the gate material itself tolerant of the dopant-activation anneal, the industry collapsed what could have been a separate masking step into the gate deposition itself, and that one architectural decision — not polysilicon's conductivity — is why the material held the gate-electrode job for forty years before leakage, not misalignment, finally forced a change. And every time that gate gets etched, the same zero-margin interface that made polysilicon the right deposition choice in 1968 is still what the endpoint algorithm is protecting today.

Take poly process further

Ask the copilot about this term, or have our engineers assess it against your process.