Electronic-Grade MTBE by SEMI Grade: G1 to G5 Metal Limits, Particle Counting and the Only Spec That Matters

Methyl tert-butyl ether (MTBE, CAS 1634-04-4) is a branched aliphatic ether with a normal boiling point of 55.2 °C at 101.3 kPa, a closed-cup flash point of -28 °C, a vapour pressure of 27 kPa at 20 °C, and a dynamic viscosity of 0.34 mPa·s at 25 °C. Electronic-grade MTBE is specified in semiconductor lithographic support operations, particularly in edge-bead remover formulations for KrF and ArF photoresist processing and in post-etch organic residue rinsing. The material contains no aromatic fraction, and its ether oxygen provides Hansen solubility parameter behaviour that promotes dissolution of novolak films without the extractable profile of aromatic hydrocarbon solvents. The same ether functionality is peroxide-forming; shipment and storage under inert gas with a stabilizer or inhibitor are mandatory for all G1 through G5 material. In a wafer-fab purchasing specification, the solvent is not qualified by a single gas-chromatographic assay but by a contamination-control bracket: Electronic-Grade MTBE by SEMI Grade: G1 to G5 Metal Limits, Particle Counting and the Only Spec That Matters.

The SEMI grade framework for organic solvents is a contamination classification, not a replacement for process-specific qualification. A raw material lot is released only after trace metal analysis, particle counting, water determination, non-volatile residue measurement, and peroxide titration have been completed on the same homogenized container. Metal limits are the most visible line item because ionic residues after spin drying are immobile on the wafer and cannot be removed by an additional rinse. Particle count is a point-of-use parameter that depends on filtration media and dispense-line conditioning. Peroxide concentration is the only specification that must be verified after every hold period rather than at incoming release alone.

What Does the G1-to-G5 Metal Grade Actually Control?

The G1 through G5 classification establishes trace metal ceilings for the cations that affect device electrical performance: sodium, potassium, magnesium, calcium, aluminium, iron, nickel, copper, zinc, lead, and often titanium, chromium, manganese, cobalt, tin, barium, and lithium. In the G1 grade, total specified metals are controlled to ≤1000 µg/kg; G2 tightens to ≤100 µg/kg; G3 tightens to ≤10 µg/kg; G4 tightens to ≤1 µg/kg; and G5 tightens to ≤0.1 µg/kg. The actual number of elements and per-element ratios differ among suppliers. In practice, the grade is read through a certificate of analysis that lists ICP-MS data and method detection limits. Sodium and potassium are limiting elements for G4 and G5 because they appear in ambient dust and are readily leached from glass surfaces. Iron, nickel, and chromium become limiting when stainless-steel components are present upstream of the final fluoropolymer filter. Copper is electrically fatal at extremely low concentration because it diffuses rapidly through silicon and introduces a deep-level recombination centre. A total metal grade may pass while an unlisted copper concentration above 0.05 µg/kg causes a yield drop. This is why device manufacturers require per-element data rather than only total metal data for G4 and G5 MTBE.

ICP-MS analysis is performed with ISO 17294-2:2016 methodology using collision/reaction cell mode for transition elements and cold plasma mode for sodium and potassium. The sample introduction system is PFA-based; borosilicate glass is excluded from the sample path. The instrument detection limit for sodium in organic solvent after dilution is typically 0.02 µg/kg, but the practical quantification limit is controlled by the laboratory blank. The grade assignment is valid only when the blank is less than 10% of the specified limit. For G5, this requires a blank below 0.01 µg/kg, which is at the edge of routine ICP-MS capability and explains why many suppliers report G5 as a filtered sublot rather than a bulk production grade.

For 130 nm and older nodes, G2 or G3 MTBE is usually adequate because the tolerable surface metal contamination is higher. For 7 nm or 5 nm logic nodes and for high-k/metal-gate integration, edge-bead remover and post-etch rinse solvents are frequently specified as G5 because the gate oxide thickness is below 2 nm and a single sodium or potassium event at the gate interface may create a threshold-voltage shift. The total metal number is a coarse screen only. A lot can pass total metals and still fail because one element exceeds its per-element limit, especially after the solvent has contacted a pump head, valve body, or filter housing that has not been pre-flushed with the same grade material.

Particle Counting: Laser Light-Scattering Response in a Low-Refractive-Index Ether

Liquid particle counting in MTBE is performed with an optical particle counter calibrated to ISO 21501-2:2019 using polystyrene latex reference spheres in an organic diluent. The primary detection threshold is ≥0.2 µm; the ≥0.1 µm threshold is reserved for incoming stability studies because the refractive index of MTBE (1.368 at 20 °C) is closer to that of the suspending medium than water, reducing the scattered-light signal from small particles. The sample is fed from a pressure-balanced PFA manifold under 20–40 kPa nitrogen headspace to suppress bubble formation, and the sample is temperature-conditioned to 20 °C ±0.5 °C before entering the detector. Without temperature control, a 2 °C cooling can create microbubbles that are counted as particles at ≥0.2 µm. A recirculating point-of-use filtration train using 0.05 µm PTFE membranes typically reduces ≥0.2 µm counts to 10 particles/mL or lower, but the count rises after filter change-out and after drum transport because the low viscosity of MTBE allows resuspended solids to remain in the liquid column longer than in a higher-viscosity solvent. The process engineer must qualify the filter train by particle-retention challenge testing with 0.2 µm latex spheres and must not assume that a clean incoming drum remains clean after transfer lines are heated or after the dispense nozzle has been idle for more than 4 h. Published data for MTBE-specific particle retention curves is limited; the specification is therefore written as a point-of-use count rather than a bulk liquid retention value.

Inter-laboratory reproducibility of particle counting in MTBE is worse than for water because the sample matrix evaporates at the detector and condenses in the exhaust line, causing pressure fluctuations. The sample cell must be cleaned with filtered solvent between runs; otherwise an organic film forms on the windows and increases background scatter. The particle counter is typically operated at a coincidence count loss below 2%, which corresponds to approximately 8000 particles/mL at ≥0.2 µm. For high-purity MTBE this limit is not approached unless the filter has failed. A single count above the action limit requires a resample rather than an immediate line shutdown, but two consecutive excursions require filter replacement and dispense-line flushing.

Particle count and metal grade are coupled through the filtration system. A PTFE filter cartridge is not inert in the first 30 min of contact; it releases fluoride and organic wetting agent. If the filter is not pre-flushed with 5–10 L of G4 or G5 MTBE, the first wafer batch receives a pulse of potassium and non-volatile residue. The flush volume is determined by conductivity and liquid particle count, not by total organic carbon alone. In high-volume lithography tracks, a filter change is commonly scheduled after 500 L of dispensed solvent or 72 h of line idle time, whichever occurs first, because peroxide decomposition products shorten the filter life and increase differential pressure.

Why Peroxide Concentration Is the Only Spec That Matters at the Dispense Point

Peroxide concentration is measured by iodometric titration according to ASTM E298-17 or by UV-visible ferrous thiocyanate complexation. The ether α-carbon reacts with dissolved oxygen under ambient light to form hydroperoxides; the rate increases with storage temperature and headspace oxygen. The peroxide concentration affects thermal safety and wafer surface contamination more than any metal line or particle count. At ≤5 mg/kg as active oxygen, MTBE is handled as a normal flammable solvent: closed dispense under nitrogen, no open evaporation, and no contact with copper or copper alloys. At 5–10 mg/kg, the material is still process-usable but requires shortened bath life and increased differential-pressure monitoring because peroxide decomposition products collect on the 0.05 µm filter. At >10 mg/kg, the solvent must not be heated, distilled, or evaporated in an open vessel; peroxide decomposition at the boiling point of 55.2 °C is exothermic and can initiate a runaway oxidation in the vapour space. A dissolved peroxide is invisible to optical particle counting and to ICP-MS. It is also not removed by filtration; it passes through the filter and remains after spin drying, where it can oxidise metal surfaces at contact vias and increase time-zero leakage current. For this reason, the only spec that matters at the dispense point is peroxide concentration, not the G5 metal label or the 0.2 µm particle count.

Material in G1 through G3 grades is normally used in non-critical cleaning operations where the peroxide limit may be relaxed to 10 mg/kg if the bath is sampled daily. For G4 and G5, the dispense system should maintain a nitrogen blanket at 20–40 kPa and the point-of-use container should be sampled for peroxide at the start of each shift and after any line idle interval longer than 4 h. Containers that have been opened and resealed more than once should not be returned to a G5 point-of-use stock because headspace oxygen replacement accelerates peroxide growth. The solvent is incompatible with copper, brass, and aluminium surfaces in pump heads and valve bodies; all wetted parts should be PFA or PTFE. Peroxide inhibitors, if used, must be selected for semiconductor compatibility and must not contribute metal or particle load; the inhibitor concentration itself becomes a non-volatile residue line item. Specifications that omit peroxide concentration, or that allow a peroxide index above 5 mg/kg at the dispense point, cannot be read as electronic-grade.