For the first time, a layer thickness sensor has been integrated into a vertical furnace for in situ sensing of the layer growth as well as for post process control of the batch. Because of its high accuracy and versatility, an in situ spectroscopic ellipsometer (SE) was selected. The adaption of the SE to the vertical furnace was performed with only minor modifications to the furnace geometry. The ellipsometer light beam is guided through the base plate into the furnace tube and directed onto the wafer by quartz glass prisms operated in total internal reflection (TIR) mode. This arrangement introduces an additional phase shift in the polarization state of the light which can be determined and subtracted from the measured phase shift. The ellipsometer setup is mechanically fixed to the boat loader. SE measurements therefore can be performed with the wafer boat out of the furnace tube as well as during the process run with the wafer boat inserted. The SE thus can be used for in situ end-point detection during layer growth and also for post process measurements on selected wafers of the batch during the unloading sequence. A realtime controller and a run-by-run controller integrated into the furnace controller utilize the layer thickness data measured in the in situ and in the post process mode, respectively, for immediate and automated correction of the parameter settings during the actual process run and for the following process run. The advantages of this novel furnace control system such as a reduction in time needed for process optimization, the avoidance of monitor wafers and a better control in integrated multilayer processing can be beneficial for future thermal batch/minibatch processing.

We have demonstrated a correlation between post-metal etch ashrate performance and yield on CMOS technology integrated circuit (IC) product wafers etched at metal etch with high polymer producing recipes. This paper presents details of high polymer forming metal etch processes and two novel techniques developed for quantifying the amount of polymer formed during etch. Data are presented which illustrate a correlation between electrical test yield, post metal etch ashrates and the amount of polymer formed during the etch. Finally, recognizing that the metal etch recipe/ashrate correlation does not explain all forms of yield loss due to leakage, we suggest other areas for investigation.

In semiconductor manufacturing, a key measure of product quality is the number of particles added during a fab processing step. Usual statistical process control methods assume an underlying normal distribution for the data. Particle measurements typically display skewed, nonnormal distributions, hence traditional control limits are not valid. This paper examines using bootstrap estimates of the upper percentiles of a distribution to find control limits for particle charts.

Particles from four stages: prior to metal sputter, after metal sputter, metal photo (ADI) and metal etching (AEI) were detected with KLA inspection tool. Different type defect counts at four stages were discussed in detail. Correlation of die yield and visual defect are analyzed. Also a technical megasonic clean prior to metal sputter which expected to reduced particles during the metal process is discussed. Experimental result shows 53.78% of the visual defects were caused prior to metal sputter, 25% occurs at metal sputter, 13.64% occurs at metal photo and 7.58% occurs at metal etching, respectively. According to the yield sorting results, 40% of visual defect which occurs prior to metal sputter, 25% at metal sputter, 83% at metal photo and 80% at metal etching are considered killer defects, respectively. Megasonic clean before metal sputter cannot reduce particle, even cause yield 3 - 8% drop due to contact resistance rising. All the evidence point out the yield impact relating particle contributed by each stage cannot be ignored.

A cycle time (CT) reduction methodology has been developed in the Lucent Technology facility (former AT&T) in Madrid, Spain. It is based on a comparison of the contribution of each process step in each technology with a target generated by a cycle time model. These targeted cycle times are obtained using capacity data of the machines processing those steps, queuing theory and theory of constrains (TOC) principles (buffers to protect bottleneck and low cycle time/inventory everywhere else). Overall efficiency equipment (OEE) like analysis is done in the machine groups with major differences between their target cycle time and real values. Comparisons between the current value of the parameters that command their capacity (process times, availability, idles, reworks, etc.) and the engineering standards are done to detect the cause of exceeding their contribution to the cycle time. Several friendly and graphical tools have been developed to track and analyze those capacity parameters. Specially important have showed to be two tools: ASAP (analysis of scheduling, arrivals and performance) and performer which analyzes interrelation problems among machines procedures and direct labor. The performer is designed for a detailed and daily analysis of an isolate machine. The extensive use of this tool by the whole labor force has demonstrated impressive results in the elimination of multiple small inefficiencies with a direct positive implications on OEE. As for ASAP, it shows the lot in process/queue for different machines at the same time. ASAP is a powerful tool to analyze the product flow management and the assigned capacity for interdependent operations like the cleaning and the oxidation/diffusion. Additional tools have been developed to track, analyze and improve the process times and the availability.

UV-transparent silicon nitride (UV-SiN) films were deposited in a plasma enhanced chemical vapor deposition (PECVD) reactor. The material and optical properties of the films were characterized for non-volatile memory devices as passivation layers. Factorial designed-of-experiments were used to study the dependence of the film properties on various process parameters. The UV-transparent SiN films are characterized by deposition rate, film uniformity, UV transmittance, film stress, refractive index, wet etch rate and hydrogen content. In this study, high quality SiN films with low compressive stress, low hydrogen content and high UV transmittance (greater than 80% for 1.5 micrometer thick film) were successfully deposited and this film has been incorporated into the passivation scheme of submicron FLASH devices. Because the absorption coefficient of silicon material is very high in the UV region, direct and accurate UV transmittance measurement for SiN by optical transmission spectrometry can only be made when SiN is deposited on quartz substrate. In this work, a simple method was used to calculate the UV-transmittance of SiN film deposited on silicon wafers from the measured extinction coefficient using a spectro-ellipsometer. This method has been further demonstrated for in-line monitoring of UV transmittance of SiN films.

In a large manufacturing environment, one of the process engineer's tasks is to ensure that all the equipment and its associated processes are well under control. This is usually achieved by running a daily monitor test on each single piece of equipment. Unfortunately, this leads to a large number of test wafer usages and becomes very time consuming if an out of control situation is detected. Things get even more complicated when the overall process is done by adding several individual processes, since the result of the test on one equipment depends on how well another one did perform. In such case, false alarms might happen and the wrong process or equipment shut down. Therefore, a need for a more reliable method must be proposed. This paper intends to describe one possible solution, and the first results of its implementation in the photolithography area.

The mechanism and the main features of the effect of anisotropic local melting of semiconductor surfaces which can be observed at definite regimes of homogeneous irradiation by powerful pulses of coherent and incoherent light are investigated. The dynamics of this effect was studied for the first time. Using rapid filming microphotos of the silicon surface were taken. On implanted silicon, the dynamics of reflection of the probing He-Ne laser radiation under heating with powerful optical irradiation is measured. The results can be explained by the different degree of semiconductor superheating with respect to the equilibrium temperature of melting. It is demonstrated that under certain conditions light pulse itself generates centers for the formation of liquid nuclei.

This paper presents a prototype etch rate sensor which correlates film thickness with the change in resonant frequency that occurs in a micromachined platform during etching. The platform is suspended over a drive electrode on the surface of the substrate and electrically excited into resonance. As material is etched from the platform, its resonant vibrational frequency shifts by an amount proportional to the amount of material remaining etched, allowing etch rate to be inferred. As a proof-of-concept experiment, a platform made of DuPont 2611 polyimide electrodes is fabricated. The sensor is driven into resonance electrostatically, and the shift in resonance is detected by monitoring the change in impedance between the drive electrode and platform as the drive frequency is swept. To enhance filtering of the sensor signal in the noisy plasma environment, the platform was designed such that the ratio of the plasma frequency to the fundamental mode of the sensor is approximately 400:1. The prototype was etched in a plasma therm RIE system in a CHF₃/O₂ plasma for 60 second intervals. After each etch, the thickness of the platform and its resonance frequency was measured, and it was confirmed that the resonant frequency decreased with decreasing film thickness. Ultimately, the sensor has the potential to allow in-situ process monitoring of both etch rate and uniformity at a nominal cost.

Defect reduction is one of the most challenging problems in the fabrication of advanced MOS devices. In-line inspection scheme of product wafers is essential for timely detection of process anomalies and providing real time feedback to the manufacturing process. A particle contamination of a specific module was observed from the inspection results on product wafers. Through partitioning studies, the tungsten (W) deposition process was identified as the particle source. This paper describes the methodology used to identify and diagnose the particle source in the W deposition. After the etch-back process, particles of abut 1 to 2 micrometers were randomly scattered on the entire wafer. The number of residue ranges from several hundred to a few thousand, forming a haze-like defect. A focused ion beam (FIB) cross section showed a hollow section of the W film, indicating improper reaction during grain growth. It was discovered that a variation in the reactant ratio during the nucleation step was causative. The change in the ratio of the reactants was in turn caused by the severe overshoot of the reactant gas at the beginning of the reaction. The pump-down step of the reactant gas prior to the flowing in of the reactant gases, which greatly enhanced the overshoot, was omitted. A two-step gas flow rate in nucleation step was developed to alleviate the effect of gas flow dynamics and thus reduced the overshoot of the gas. With these modifications and other proper corrective actions implemented, the residue was completely eliminated.

Systems engineering methods can be applied to almost any engineering task, this paper outlines the approach adopted to optimize maintenance activities in a wafer fab environment with applications on ion implanters and sputtering tools. The trade-off was to design an effective maintenance system keeping in mind the production requirements. The analysis phase identified areas of improvements in preventive maintenance (PM) scheduling, corrective maintenance (CM) control and reporting, logistic support management and equipment maintainability. All those tasks have been included in the work breakdown structure (WBS) and implemented. The results were a strong increase in equipment availability in the implanter's case and a reduction on failure rate for sputters. Future developments will consider the introduction of expert systems to help personnel in diagnoses and machine status control.

The semiconductor manufacturing industry is continually riding the edge of technology as it tries to push toward higher design limits. Mature fabs must cut operating costs while increasing productivity to remain profitable and cannot justify large capital expenditures to improve productivity. Thus, they must push current tool production capabilities to cut manufacturing costs and remain viable. Working to continuously improve mature production methods requires innovation. Furthermore, testing and successful implementation of these ideas into modern production environments require both supporting technical data and commitment from those working with the process daily. At AMD, natural work groups (NWGs) composed of operators, technicians, engineers, and supervisors collaborate to foster innovative thinking and secure commitment. Recently, an AMD NWG improved equipment cycle time on the Genus tungsten silicide (WSi) deposition system. The team used total productive manufacturing (TPM) to identify areas for process improvement. Improved in-line equipment monitoring was achieved by constructing a real time overall equipment effectiveness (OEE) calculator which tracked equipment down, idle, qualification, and production times. In-line monitoring results indicated that qualification time associated with slow Inspex turn-around time and machine downtime associated with manual cleans contributed greatly to reduced availability. Qualification time was reduced by 75% by implementing a new Inspex monitor pre-staging technique. Downtime associated with manual cleans was reduced by implementing an in-situ plasma etch back to extend the time between manual cleans. A designed experiment was used to optimize the process. Time between 18 hour manual cleans has been improved from every 250 to every 1500 cycles. Moreover defect density realized a 3X improvement. Overall, the team achieved a 35% increase in tool availability. This paper details the above strategies and accomplishments.

As device linewidths shrink to 0.5 micrometer and below, the ECD (electrical critical dimension) technique is a good choice for in-line CD (critical dimension) measurements on conducting substrates. In this size regime, ECD is poised to replace SEM (scanning electron microscope) as the standard tool of the semiconductor industry, with a measurement capability significantly better than that of the CD SEM. Keithley Instruments has developed an advanced electrical prober for use at Motorola's MOS-12 facility. This tool provides in-line CD measurements in a completely automated, hands-off 'load-and-go' mode that requires only wafer loading, measurement recipe loading, and a 'run' command for processing. Its expected capability will permit measurement of lines thinner than 0.5 micrometer accurately, and with complete data transfer to the factory data collection and analysis system. The measurement results are well correlated to SEM top-down measurements and their use as a SEM replacement would not result in any loss of necessary process characterization information. Results are also included from a contamination study performed as part of the production qualification process for this tool. Particulate contamination need not be a concern, even for In-Line testing. Conversion from SEM CD to ECD measurements on metal and polysilicon layers is expected to provide a number of benefits, including better process control for very thin lines through more data with improved repeatability and stability. Other expected benefits include equipment cost savings and improved cycle time.

An inexpensive automated DLTS system has been developed in modular form consisting of modules such as a capacitance meter, pulse generator, DLTS system timing controller, data acquisition system, PID temperature controller, cryostat with LN₂ flow control facility, etc. These modules, except the capacitance meter and pulse generator, have been designed and fabricated in the laboratory. Further they are integrated and interfaced to PC AT/386 computer. Software has been developed to run the spectrometer, collect data and off-line data processing for the deep level parameters such as activation energy, capture cross-section and density. The system has been used to study the deep levels of platinum in n-type silicon and their passivation by atomic hydrogen. The estimated activation energy of the two acceptor levels are E_c-0.280 eV and E_c-0.522 eV and their capture cross sections are 2.2E-15 cm^-2 and 4.3E-15 cm^-2 respectively. These levels are found to be reactivated when the hydrogenated samples are annealed in the temperature range 350 - 500 degrees Celsius. The mechanism of passivation and reactivation of these levels are discussed.

This paper is concerned with the method of surface charge profiling (SCP) developed for in-line monitoring of front- end processes in semiconductor manufacturing. In this study a commercial SCP system is used to monitor wafer cleans in terms of oxide/hydrogen coverage of Si surfaces following cleaning with emphasis on HF last cleaning sequences. Moreover, metal contamination of bare silicon surfaces and deactivation of boron dopant in the near-surface region of p-type Si wafers are monitored. Finally, the unique capability of SCP in monitoring time-dependent evolution of characteristics of Si surfaces exposed to various ambients is demonstrated.

An investigation was undertaken to evaluate the potential of using mutual inductance (eddy current) measurements to determine doping levels in polycrystalline silicon (polysilicon). The main advantage of eddy current measurements is that they are non-contact and the oxide which grows on the polysilicon is transparent to the magnetic fields used to generate the eddy currents. This allows the doped polysilicon sample to be measured immediately following the doping process and eliminates the need to etch the oxide which must be removed for the traditional four-point probe measurement. The elimination of the HF acid etch step not only saves on materials costs, it removes a potentially hazardous step from the process and does away with the need to dispose of a hazardous waste. This paper correlates the eddy current measurements with traditional four-point probe measurements, and investigates the influences of the substrate and backside polysilicon film sheet resistances on the frontside measurement.

The paper presents new measuring technology based on slow- wave structures (SWS) application. It shows peculiarities and advantages of sensors on SWS. The method is based on the interaction between the electromagnetic wave exited in SWS and the object to be checked up. It deals as well with a transducer for non-contact measurements of surface conductivity and thickness of thin films and conductive layers on dielectric or semiconductor substrates. It shows perspectives of designing other transducers on SWS for measuring different physical parameters.

Ultraviolet (UV) thickness measurement equipment has little room for variation when determining ultra thin films which are 70 angstroms or less. High lamp performance is critical for measurement validity. A quality conscious semiconductor must have data to verify a vendor claim of 'The lamp performance will perform with no degradation for up to (xxx) hours of normal operation.' In this article we review a real case where data was collected and examined to answer important questions about lamp performance in UV measurement equipment. How long could a lamp be used before performance degraded enough to necessitate a lamp replacement? This article will illustrate how we used standards and actual measurements to collect data for this study. Plots are included showing actual collected data followed by a discussion of alternative methods for statistical examination of the data. This discussion will include an illustration of an original and useful statistical approach for determining the point in time when degradation is noticeable. The method for examining data begins with a well known but not too frequency used concept known as piecewise linear regression with a fixed point of join. Then we enhance the method by turning the join point into a variable that is 'floated' using an iterative non-linear regression approach.

Individual measurements and moving-range control charts are applied when the quality characteristic follows a normal distribution. In the semiconductor industry, certain quality characteristics are monitored where the data do not follow a normal distribution. In this paper we consider alternative design strategies for control charts where the distribution of the data is skewed. Although we use skewed data, these techniques are also applicable to other non-normally distributed data. We discuss using a 'state of the art' curve-fitting method, data transformations, and nonsymmetrical limits for the control charting of skewed data. This paper provides alternatives for the design, implementation, and maintenance of process control for non- normal or skewed data.

SVG/Thermco 7000 series VTRs (vertical thermal reactors) were purchased capable of running 100 production wafers in a single run. Manufacturing needs required an increase to 125 production wafers per run. Polysilicon deposition furnaces with five zone temperature control were capable of running 125 production wafers, but furnaces with three zone temperature control were not. The three zone temperature controlled furnace showed an elevated thickness signature near the top test wafer position. This increased thickness measured near upper specification limits. A designed experiment was run based on dummy wafer placement and end zone temperature setpoints. The optimized dummy wafer load and top and bottom temperature setpoints improved the wafer within-run uniformity from 8.0% to 1.8%.

Effective statistical process control (SPC) necessitates the use of the proper control chart, used to monitor and control the important process characteristics. The control chart can be used to monitor stability for a process which is normally distributed. Scanning electron microscopy (SEM) critical dimension (CD) measurement, however, shows a systematic variation in repeated measurements of a sample in addition to the Gaussian variation, due to the nature of electron beam irradiation of materials. Because of this systematic variation the standard control charts do not work well for the control of this process. Addition of a slope-subtraction algorithm to the data collection system provides display of the slope-corrected data in addition to the raw data display, and permits the stability of the metrology tool to be demonstrated.

A practical automatic feedback system of exposure energy is proposed to improve the polygate photo critical dimension (CD) control. By the curve of (Delta) (CD) versus (Delta) (exposure energy), we can give the right exposure energy to the current lot by the preceding 5 (or 3) lots' CD, measured by CD SEM. To be easily used, the curve is approached as a linear function. Two months experiment of exposure energy feedback system is carried out on actual 0.5 micrometer i- line SRAM process. The capability index (C_pk) of polygate photo CD has been dramatically improved.

Experiments were carried out to characterize and reduce defectivity at interlayer dielectric (ILD) deposition on a chemical vapor deposition (CVD) reactor. There was evidence that some of the yield limiting defects for product originated from ILD deposition. Several approaches were taken to address this problem. Film thickness dependence, plasma extinguishing techniques, and in-situ cleans were investigated. Film thickness was a concern because a new ILD process requiring a thicker film had been implemented in conjunction with chemical mechanical polishing (CMP). It was found that particle size tends to increase with increasing film thickness. A literature search disclosed a recommended procedure for reducing particles based on a plasma extinguishing technique that effectively levitated particles before flushing them from the chamber. This technique was evaluated with inconclusive results. Through the course of the study it was found that the in-situ clean times had the largest effect on defectivity. Non-optimized in-situ cleans caused higher particle counts. An internally manufactured optical endpoint detector was used to optimize cleans. Recent work compares the traditional two-step in-situ clean to a single-step clean. Two equipment configurations and clean gas combinations were tested. Results document improvements in particle performance and throughput.

Upstream control and in-situ defect monitoring have been successfully implemented to track and improve particle performance for CVD tungsten films. An in-situ 'dark field' particle sensor was used to monitor process defect levels during tungsten deposition. The sensor was mounted directly below the CVD tungsten deposition chamber on the foreline. Information from the in-situ particle sensor and the CVD system process parameter analog output were routed to a stand alone computer where the incoming signals were recorded. In-situ particle and process parameter signals were monitored using the standard production recipe for tungsten deposition. From this data the internal variables (process parameters) affecting in-situ particle levels were determined. The internal factors were varied over a selected operating range and the in-situ particle levels quantified for each factor. These results were used to define the CVD tungsten process operating window in terms of in-situ defectivity. Additionally, other external sources were identified as contributors to in-situ defectivity.

In the modern semiconductor processing facility CD (critical dimension) measurements on 200-mm diameter wafers proceed in a fully automated mode requiring only wafer loading, measurement recipe loading, and a 'run' command for processing. To provide effective metrology support to this facility the scope of CD measurement automation has been extended to the entire metrology scheme, from job file generation to CD measurement and data collection. Upon completion of all sample measurements, data is uploaded to the factory's data collection software system via a SECS II interface, eliminating the requirement of manual data entry. This paper presents the methodology used to perform calibration of the automated CD metrology scanning electron microscope (SEM) for sub-0.5 micrometer process control. A commercially available SEM pitch standard, traceable to JQA (Japan Quality Assurance) has been characterized and incorporated into a standard, automated calibration procedure performed as part of the regular tool qualification. This standard contains lines with pitch on order of 0.240 micrometer over a 4 mm² area and is useful at low accelerating voltage. Since the low accelerating voltage CD measurement SEM has assumed an important role in modern semiconductor process control, we investigate the use and performance of the standard in a representative instrument. Evaluation of this reference by SEM CD measurements shows suitable performance for submicron dimension calibration.

Industrial engineering is fairly new to the semiconductor industry, though the awareness to its importance has increased in recent years. The U.S. semiconductor industry in particular has come to the realization that in order to remain competitive in the global market it must take the lead not only in product development but also in manufacturing. Industrial engineering techniques offer one of the most effective strategies for achieving manufacturing excellence. Industrial engineers play an important role in the success of the manufacturing facility. This paper defines the industrial engineers role in the IC facility, sets the visions of excellence in semiconductor manufacturing and highlights 10 roadblocks on the journey towards manufacturing excellence.