Reactor overview with detail of quantified dosing section. See text and Table I for description of fill, measurement, and dose and purge functions. Dose quantification and control occurs in real time. All valves in this figure are rapid action, with timing controlled by computer. High vacuum conductance liquid nitrogen-filled trap prevents ALD reactants from entering pump and oil from backstreaming into reactor.
Reaction chamber and pumping section detail. Gate valve 1 and the system vacuum valve are for high vacuum and viscous flow type operation, respectively. The quadrupole mass spectrometer (QMS) is differentially pumped by turbopump 2 to a pressure of ∼10−8 Torr, as depicted with a crosshatch pattern, through a removable internal vacuum flange with a pinhole. Not shown are the QCMs, capacitance manometers close to the substrate, and precision leak valve.
Control of dose amount achieved from drawing vapor off of a liquid TMA source, with fill pressures (left axis) and dose amounts (right axis) calculated using calibrated dose volume and fixed vessel temperature of 100 °C. Upper curve represents room temperature source and lower curve with source cooled to 0 °C. Points are averages of 30 measurements, error bars are ± 1 standard deviation of the 30 measurements and are primarily to illustrate reproducibility of fill. Inset zoom is of lower left boxed region of lower curve.
Example of method for calculation of integrated impingement flux (upper half) and for quantification of dose (lower half). (a) Function fit of a single system pressure peak to form of Eq. (13) . Top plot is the residual, middle plot is the data and fit together; bottom graph is the fit function alone. (b) Illustration of the fits of 5 such pressure peaks. (c) Quasi-static doser pressures controlled with fill time. The rise is during the fill, the decay is the emptying of the dose into the reactor. (d) Doser fill consistency example. Circles represent the locations of peak pressure and hence total dose vessel fill pressure because at the peak, all valves enclosing the dose vessel are closed momentarily. Using the fixed volume and temperature of the dose vessel, true absolute doses are converted to micromoles using the ideal gas law. All doses shown are of TMA vapor drawn from a 0 °C source into the dose vessel, held at 100 °C. Fills and empties of both the reactor (top half) and dose vessel (bottom half) are into high vacuum (P < 10−5 Torr); negligible compared to the quantities measured.
Relationship between the net exposure, measured as the integrated molecular impingement flux (Eq. (6) ) and the quantified dose in micromoles. Each point represents the average of 5 integrated impingement fluxes and 5 dose amounts per fill time, as depicted in Figs. 4(b) and 4(d) . The straight line is a least-squares fit to the data. The number of equivalent TMA monolayers that correspond to a given net exposure is plotted on the right axis. This is determined by dividing the net exposure by the estimated monolayer packing density of TMA, 3.79 × 1014 molecules per cm2 (see text). As an aside: note that the actual number of equivalent TMA monolayers of flux at the QCM and substrate is smaller than this number during a deposition because significant surface area exists along the flow path prior to the substrate, and this reactive surface area would consume a portion of the quantified dose before reaching the substrate. This upstream surface area was extensively saturated with TMA immediately prior to the measurements in Figure 4 used in this correlation.
Dynamics of differently sized doses into high vacuum flow, without carrier gas. Dose size increases with fill time as does peak reactor pressure resulting from the dose. An estimate of the true duration of a dose's exposure to the substrate, the full width at half maximum (FHWM) of each pressure peak, is displayed in the inset, demonstrating the inverse relationship between dose size and dose duration. Note this effect will be enhanced for larger ΔP as the dose travels through the reactor and would not necessarily follow the same relationship for continuum flow situations (see text).
(a) Mass spectrum obtained for trimethylaluminum (TMA). (b) Calibration curve of actual reactor pressure of pure TMA vapor to QMS signal in counts per second at mass to charge ratio 57, with linear fit. This m/z value is the most abundant Al-containing ion. Larger signals at m/z = 15 (CH3+) and m/z = 16 (CH4+) overlap with the CH4 product of reaction, so would be poor choices for a TMA pressure calibration.
(a) Monitoring m/z = 57 for TMA, m/z = 18 for water, and m/z = 15 and 16 during the course of 6 typical ALD cycles for Al2O3 growth. Because m/z = 15 and 16 are associated with methane, contributions at each peak during a TMA dose may be due to the fragmentation of TMA molecules in the ionizer of the QMS, or the product of reaction with walls of the reactor, or even possibly CVD-like reactions if the water purge step was insufficient. During the water dose, they would most likely be associated with releasing ligands from surface bound TMA molecule fragments. (b) Four typical ALD cycles showing the QCMs masses (left axis) correlated to the QMS peaks from TMA and water (right axis). Total calibration of all of the QMS partial pressures, combined with quantified dosing and the QCMs, allows for accurately accounting of where and when the reactant is consumed, and byproducts of the reaction produced (in progress as of this writing).
Individual mass spectrometer calibrations for pure methane alone, and for the apparent methane coming from TMA alone. Note that the m/z = 15 species is greater in magnitude than m/z = 16 resulting from TMA, but the opposite trend is the case for pure CH4. This is clearly due to the fragmentation of TMA in the ionizer of the mass spectrometer. The difference in slopes, however, means that the species at these m/z ratios may not easily be discriminated as coming from unreacted TMA or from CH4 reaction product entering the QMS (see text).
Parallel mass measurement onto two QCM crystals. Experiment 1 was 50 Cycles of Al2O3 growth from TMA/water at 70 °C onto two crystals previously coated with Al2O3 or “uncoated.” This was meant to serve as a control. During Experiment 2 the same 50 cycle sequence was performed with a stearic acid self-assembled monolayer (SAM) on crystal 2, and with only bare or uncoated Al2O3 on crystal 1. The lower plot shows substrate-inhibited growth, yet the TMA/water reaction is not completely inhibited even on this hydrophobic substrate (see text).
Relative reactive sticking probability (left axis) and mass per cycle (MPC—right axis) as a function of cycle number. The MPC measurement shows how many cycles are deposited onto the SAM-coated crystal before the same growth rate reappears as on the uncoated crystal. Presumably the SAM is completely covered by Al2O3 at that point, which also is indicated by the relative sticking probability returning to ∼1. Measurements of this type should lead to a more quantitative understanding of substrate (or surface-chemistry-based) selective growth. TMA is considered an especially reactive precursor and is unlikely to grow completely selectively, yet clearly exhibits some nucleation delay on a methyl-terminated SAM.
Typical viscous flow operation of the reactor.
Article metrics loading...
Full text loading...