Search

One of the bottlenecks to building semiconductor chips is the increasing cost required to develop chemical plasma processes that form the transistors and memory storage cells1,2. These processes are still developed manually using highly trained engineers searching for a combination of tool parameters that produces an acceptable result on the silicon wafer3. The challenge for computer algorithms is the availability of limited experimental data owing to the high cost of acquisition, making it difficult to form a predictive model with accuracy to the atomic scale. Here we study Bayesian optimization algorithms to investigate how artificial intelligence (AI) might decrease the cost of developing complex semiconductor chip processes. In particular, we create a controlled virtual process game to systematically benchmark the performance of humans and computers for the design of a semiconductor fabrication process. We find that human engineers excel in the early stages of development, whereas the algorithms are far more cost-efficient near the tight tolerances of the target. Furthermore, we show that a strategy using both human designers with high expertise and algorithms in a human first–computer last strategy can reduce the cost-to-target by half compared with only human designers. Finally, we highlight cultural challenges in partnering humans with computers that need to be addressed when introducing artificial intelligence in developing semiconductor processes.

The ever-shrinking dimensions of microelectronic devices has mandated the use of plasma processing in integrated circuit (IC) factories worldwide. Today the plasma-processing industry has grown to over $3 billion in revenues per year, well in excess of predictions made only a few years ago. Plasma etching and deposition systems are also found throughout flat-panel-display (FPD) factories despite the much larger dimensions of thin-film transistors (TFTs) that are used to switch picture elements (pixels) on and off. Besides the use of plasma in etching and depositing thin films, other processes include the following: removal of photoresist remnants after development (descumming), stripping developed photoresist after pattern transfer (ashing), and passivating defects in polycrystalline material. Why are plasma processes so prevalent?In etching, plasmas are used for high-fidelity transfer of the photolithographically defined pattern that defines the device or circuit. More generally, plasma provides the means to taper sidewalls. In Si processing, the sidewalls must be nearly vertical to obtain high density integration and faster performance. However in making FPDs, sidewalls are tapered to obtain uniform step coverage and reduce shorting. In deposition, plasmas are used to enable processing at low temperature. For both etching and deposition, only plasma processing provides an economically viable means for processing large area substrates: 300 mm for Si and more than 550 × 650 mm for FPDs. It is the ability to scale uniform reactant generation to larger areas that sets plasma apart from beam-based processes that might otherwise offer the desired materials modifications. The nonequilibrium characteristics of plasma further distinguish this processing method. Energetic electrons break apart reactant precursors while ions bombard the surface anisotropically.

Dry etching of compound semiconductors is becoming increasingly important as design ruler shrink for electronic devices. For photonic device applications, dry or plasma etching is used fin- device isolation, fine-line pattern transfer, and fabrication of optical quality interfaces. As has been well established for Si and W. plasma etching at reduced temperatures can provide superior critical dimension control and obviate the need for operating at high bias voltages that produce excessively energetic ion bombardment t. In this work, we explore low-temperature (−60 C to +60 C) etching of the compound semiconductors GaAs, AlGaAs, and AlAs, In addition to improving etch anisotropy, which provides critical dimension control, rye find thut processing at lower temperatures improves microuniformity and reduces loading effects. At high lemperaturcs, where larger samples are observed to etch more slowly than smaller pieces (loading effect), etching rates appear limited bv reactant transport to the wafer. In this regime, both microuniformity and macrouniformity arc poor. As the temperature is reduced, the etching rate becomes limited by surfitce processes us a residue containing the semiconductor elements, etchant gases, and residual background gases forms on the surface. hi this regime, the etch rare becomes independent of surface area and uniformity is improved.

We evaluated the characteristics of a-Si:H/a-SiNx:H thin film transistors (TFTs), and of separately deposited a-Si:H films, as functions of the a-Si:H deposition power in a high-rate, large-area, 40 MHz PE-CVD system. TFT performance and a-Si:H film properties improve with decreasing power density and deposition rate. However, low defect density a-Si:H material was deposited at rates as high as 1500 Å/min. TFTs with gate nitride deposited at 1000 A/min show excellent I-V characteristics when the a-Si:H deposition power is low enough. The TFT electron mobility in the linear regime correlates well with the Urbach energy of the a-Si:H films, suggesting that the quality of the a-Si: H controls the performance of our TFTs.

Ground-state atomic hydrogen produced in radio-frequency plasma discharges (20 KHz-5 MHz) has been detectedin situ using two-photon absorption laser-induced fluorescence (TALIF). Atomic ground-state concentration measurements have demonstrated excellent spatial resolution in the interelectrode gap of a planar discharge configuration with 10 nsec temporal resolution at all phases of the RF driving voltage waveform. Concentrations were measured in gas mixtures of helium and hydrogen down to 2×1013 H atoms/cm3.

We describe the use of laser-induced fluorescence and optical emission spectroscopy to study the chemistry of CCl4 discharges. Spatially and spectrally resolved plasma-induced emission from six different species—CCl, C, Cl, Cl+, N2, and N 2 + —were contrasted with the CCl free radical concentration profile, determined from spatially resolved laser-induced fluorescence, to arrive at a simple model for the CCl4 discharge chemistry. Electrons appear to be produced at the electrode surfaces by secondary emission resulting from high-energy ion impact. As the electrons are accelerated across the sheath, they are rapidly attenuated by resonant dissociative attachment to CCl4 and its plasma-generated products. This mode of decomposition produces bright emission nearest to the electrode surfaces and is conducive to the formation of “glow polymer.” To test the validity of using N2 as an actinometer in determining the CCl radical concentration, we compared laser-induced fluorescence measurements of the CCl concentration with plasma-induced CCl emission normalized to N2 emission.

We describe the use of time-resolved laser-induced fluorescence (TRLIF) and plasma-induced emission (PIE) spectroscopy in studying the dynamics of ion transport, formation, and loss in low frequency RF plasmas, used in plasma etching and deposition. N2+ and Cl2+ ions formed in N2, Cl2, and N2/Cl2 discharges were monitored as a function of both position between the electrodes and magnitude of the applied rf potential. In the discharge center, TRLIF was used to measure ground state ionic lifetimes. In N2/Cl2 mixtures, N2+ was found to charge exchange rapidly with Cl2 and Cl to form Cl2+ and Cl+. In the electrode sheaths, the ion response to the applied potential was evident from periodic depletion of the ion concentration as a result of acceleration by the field. From the spatial variation in the ion concentration time dependence, we deduce that the sheaths expand and contract with the same period as the applied potential.

A one-dimensional fluid model is presented for time-dependent ion concentrations in chlorine containing discharges. The ion formation rate is determined from experimental plasma-induced emission intensities. The local field is estimated from spectrally resolved laser-induced fluorescence data obtained in similar, BCl3 discharges. Quantitative agreement with experiment is obtained implying that the basic assumptions in the model are valid: (1) ground state ions in the sheath are formed predominantly by electron-impact ionization; (2) Cl2+ ion motion is mobilitylimited by charge exchange collisions with Cl2 neutrals; (3) the degree of Cl2 dissociation is ∼. 65% at a power density of 1.8 W cm−3; (4) the ion concentration near the electrode is spatially uniform throughout most of the rf cycle; and (5) there are two times during a low frequency rf cycle when ions experience a strong extraction force.Semi-empirical models like the one described here should be useful in computational optimization of plasma processes such as etching and deposition, which are in wicre-spread use throughout the microelectronics industry.

We report a new technique for the in situ, non-intrusive, and sensitive measurement of plasma electric fields with high spatial resolution. Fields as small as 40 V/cm can be characterized by spectrally resolving laser-induced fluorescence from Stark-mixed parity levels. The technique is demonstrated by exciting the A1 11−X 1Σ+ band system of BC1, produced in an rf discharge through BCI3. The characterization and absolute measurement of sheath fields, which have been elusive, are now possible. These fields are preeminent in governing charged particle transport to and between electrode surfaces and in maintaining the glow discharge.

An obstacle to better design and control of plasma processes is our limited ability to monitor surface properties in situ. We describe a new plasma diagnostic technique, photoemission optogalvanic spectroscopy, which can be used to monitor surface contamination, etching rates, and end points. The experiment consists of irradiating a surface with a pulsed excimer or dye laser whose photon energy is above the substrate work function. Photoemission of electrons from the surface is detected by monitoring changes in the discharge current (optogalvanic effect). The technique is characterized by measuring signal strengths as a function of laser intensity, laser wavelength, and bias voltage in vacuum and in an Ar plasma. The technique is used to monitor contamination and etching end points in F-containing plasmas.