1. Water follows a simple empirical law
For all vacuum metals — stainless, copper, aluminium, titanium — room-temperature water outgassing is about 3×10⁻⁹/t (hours) in mbar·L/(s·cm²). Ten times the pumping time gives one order lower outgassing and pressure. Two consequences: pumping longer is not folklore, and a cleaned chamber must be pumped immediately — a re-saturated water layer wastes the pumping already done.
2. Baking fixes water, not hydrogen
Measured after 10 h pumping on stainless (Torr·L/(s·cm²)): H₂O from 2×10⁻¹⁰ to 1×10⁻¹⁴, four orders; CO from 5×10⁻¹² to 1×10⁻¹⁴; H₂ from 7×10⁻¹² to 5×10⁻¹³, barely one order and then flat. Hydrogen needs vacuum annealing or long pumping — the most commonly misread point when choosing a process route.
3. Baking temperature follows the material
Copper and aluminium reach below 10⁻¹³ with 150–200 °C for 24 h; stainless needs above 200 °C, up to 300 °C, and millimetre-thick walls may need repeated bakes. In-situ baking above 120 °C for at least 12 h is effective on metal, with 150–250 °C for 24 h as common practice, mainly for water. Viton is limited to about 200 °C; overheating deforms or decomposes it into a contamination source. This is why aluminium chambers bake easily and steel ones do not.
4. Three criteria for ending a bake
Pressure stops falling at temperature; pressure returns to target after cooling; and the mass-18 (H₂O) peak drops below mass 28 (CO/N₂) on the RGA. The first two are gauge readings; the third requires an RGA. Without one, calling a bake complete is guesswork.
5. Cleaning has three layers, not one wipe
Grease, fingerprints and loose particles (solvents, alkaline detergent, ultrasound); oxide and contaminated films (acid, electropolish or chemical polish); and the damaged layer (vacuum annealing, ion bombardment). Skipping the third is the most common domestic shortfall: the surface looks clean, but hydrogen and impurities in the damaged layer bleed out, pressure stalls at 10⁻⁷ and baking cannot rescue it. A single bare-hand contact can outgas enough to cancel hours of baking.
6. One O-ring costs two orders
A baked all-metal system reaches 10⁻¹⁰–10⁻¹¹ mbar ultimate; a single elastomer seal drops the ultimate to the 10⁻⁸ range. That is the quantitative basis for all-metal, CF-knife-edge construction in UHV — not a matter of taste.
7. Check the vintage of any particle number
ITRS 2003 (FEP Table 70a): critical particle diameter 50 nm with 59 per wafer in 2003, 40 nm with 97 in 2005, 32.5 nm with 80 in 2007, 25 nm with 68 in 2009; critical GOI surface metals ≤0.5×10¹⁰ atoms/cm², other critical metals ≤1×10¹⁰, surface carbon 1.8×10¹³, mobile ions 1.8×10¹⁰. This is the 2003 roadmap and today's limits are tighter: the value lies in units, magnitude and definition, not as a current standard.
8. Counter-intuitive: handling is dirtier than the chamber
ITRS 2003 Table 109a (defects/m², 54 nm equivalent): plasma etch 1049, CVD insulator 854, metal PVD 591, wet bench 474, CMP clean 397, while wafer handling is only 33, falling to 6 by 2009. Robots, lift pins and slit valves contribute more particles than the chamber itself — watching only the inner wall aims at the wrong target.
9. 4 Å and Ra 0.1 µm are not the same thing
ITRS specifies LVGX RMS 4 Å (about 0.4 nm) on the wafer; domestic electropolishing commonly quotes Ra 0.1–0.2 µm, an industry rule of thumb. Two orders apart, and different definitions (RMS versus Ra). Neither equate electropolish Ra with wafer-level requirements nor condemn a chamber on Ra alone: gas flow, flow field and handling sit in between.
10. Four rules that can be followed as written
Pump immediately after cleaning, never overnight in the shop; set bake temperature by material and keep Viton below 200 °C; end the bake on the three criteria, one of which must be the RGA 18/28 comparison; finish with glow discharge cleaning where residual surface contamination matters. Cleaning is a process with criteria, not an action; an action without a criterion has not been performed.
Outgassing and particles are separate problems: one is surface chemistry, the other is flow field and moving parts. Both need criteria, and criteria need to be reproducible.
FAQ
腔体洗完还是抽不下去，先查什么？
先区分出气与漏。水汽出气率按 3×10⁻⁹/t（小时）衰减，清洗后表面水层若已重新吸饱，之前抽的时间就白费了。
烘烤能解决多少问题？
250 ℃ × 24 h 可把水汽压 4 个数量级（2×10⁻¹⁰ → 1×10⁻¹⁴），但氢只压 1 个数量级，氢要靠真空退火或长时间抽气。
颗粒要求怎么换算到每片晶圆？
用关键颗粒直径与允许缺陷数折算：ITRS 2003 给出的量级是 25–50 nm 对应 59–97 颗/片，现行门槛更严，以客户规范为准。