Clean cycle prediction in etching processes via direct plasma control
03/01/2001
Roisin Cheshire, Scientific Systems Ltd., Dublin, Ireland
John Scanlan, Scientific Systems USA, San Jose, California
Gary Skinner, Altis Semiconductor, Essonnes, France
overview
Plasma processes such as etching are typically controlled with machine parameters, but direct real-time monitoring of the plasma provides added insight and control. Prediction of when a chamber needs to be cleaned is an especially good application, preventing automatic cleaning steps when they are not yet needed. An ion flux probe is one such monitor, and its functioning and a case history are described here.
With increased competition in the chip industry, DRAM, logic, ASIC, and other semiconductor products become more like commodities every day. The resulting lower margins mean that inefficient manufacturing processes are no longer affordable. Comparing the semiconductor business to more mature manufacturing industries — such as the automobile industry — reveals gaping holes in silicon processing methods. What other industry besides silicon processing goes as far as stopping a good process to check that the process is still good?
 Figure 1. The ion flux probe from Scientific Systems. |
Although the industry has defined schemes such as run-to-run control and fault identification well, the majority of manufacturing houses run process control in the time-honored fashion of machine control, with the assumption that the process will be in control if the process inputs are controlled. Process engineers continue to rely on machine-state parameters, such as gas flows and reactor pressure, for process control. The industry's increasing demands for rapid production ramps, increased throughput and yield, and low nonproduct wafer (NPW) usage press the need for a more direct approach to process control and development.
Process input measurement
Process control is often based on the process inputs, but this is not the best approach in plasma-enhanced processes, for example, where process output is a more direct function of plasma events than of process inputs. This is especially significant when we consider that many of the major steps in chip manufacturing, including etch, physical vapor deposition (PVD), and plasma-enhanced chemical vapor deposition (PECVD), are plasma-enhanced processes. While the secondary parameters defined do affect the plasma significantly, they represent only some of the inputs to the plasma, and even the best measurement and control equipment cannot measure all process inputs. Measurement and control of the plasma itself avoids concerns about inaccurate or incomplete methods of measuring a huge combination of inputs. Controlling the secondary parameters remains important, but it represents an incomplete picture.
In situ process control, in which the plasma itself is measured, is only now becoming an option with the recent development of direct plasma sensors and gauges. In the interim, optical techniques have provided some limited solutions. Commonly used for end-point detection in etch processes, optical techniques can also work for plasma to an extent. Optical techniques pick up relatively gross changes to the plasma wall, but they may not always provide the sensitivity needed to identify small process changes. Optical techniques also generally record only the chemistry, not electrical parameters, and they do not look at the plasma sheath.
Process control through plasma control
In contrast, direct measurement of the ion flux across the plasma sheath carries a strong correlation to what happens on the wafer. To address this, Scientific Systems developed the ion flux probe, which allows implementation of statistical process control on a fundamental plasma parameter. Based on patented technology, the ion flux probe operates in almost all plasma processes, including PECVD, PVD, and etch. By measuring the plasma ion flux at the reactor chamber wall during a process, the ion flux probe monitors real plasma conditions and, by direct inference, wafer processing events.
The monitor is installed much like a pressure gauge. The pressure gauge on every plasma chamber today measures the neutral gas pressure, but since what happens on the wafer is a function of the ionized gas — not of the neutral gas — measuring ion flux offers direct correlation to the process. A pressure gauge provides only an indirect measure of the ions. The ion flux probe, by measuring ion flux directly, is to plasma what the pressure gauge is to the neutral gas.
The ion flux probe plugs onto the wall of the plasma process chamber on standard access ports. Because the signal from the probe is analog, the output can be plugged directly onto the tool or read using a standard voltmeter. An optional PCMCIA card and software system allow data retrieval from a laptop PC. Integration into existing hardware and software eliminates problems typically associated with stand-alone instrumentation.
Measuring the ion flux
The ion flux probe works as a planar electrode that mounts flush with the reactor wall, measuring the ion flux with negligible disturbance to the plasma. A capacitively coupled (as opposed to DC-coupled) measurement provides an insulator effect and allows operation under heavy polymer deposition. This allows the system to operate effectively in nearly all process conditions. Figure 1 shows the probe's sensor head and control unit.
 Figure 2. Ion flux measured during a single-wafer polysilicon etch process. |
The ion flux probe electrode is biased using capacitively coupled pulsed RF power (10MHz, 2W). Because the electrode is capacitively coupled, no DC current can flow, and the electrode attains a negative bias (Vbias). This bias develops because the electron and ion fluxes must be equal to a capacitively coupled electrode. Electrons, with a much smaller mass, have a larger mobility than the ions for the same electric field strength. The DC bias then balances the ion and electron fluxes.
 Figure 3. Plasma ion flux vs. upper power and process pressure. |
When the RF pulse is switched off, the sensor electrode discharges from Vbias to the floating potential over time. If Vbias is large enough, the initial electron current is virtually zero. The ion flux probe determines the ion flux by measuring the image current returning to the capacitor while Vbias is still large enough to shield the electrode from the electron flux.
Typical use of the ion flux probe is shown in Fig. 2. In this polysilicon etch process, all measurements correspond to the plasma ions rather than to secondary process inputs.
Case study: Increasing mean time between cleans
Process engineers at the Altis Semiconductor (Infineon/IBM joint venture) plant in Essonnes, France, wanted to increase overall equipment effectiveness (OEE) by better understanding and improving dry-clean efficiency. Operating with a mean time between wet cleans (MTBC) of approximately 100 RF hours, engineers installed a Scientific Systems ion flux probe. In less than three months, the MTBC was increased to 400 RF hours.
Altis Semiconductor uses LRC9400 polysilicon etch tools for critical polysilicon gate etch process steps. The chemistry in these processes results in SixOy formation, which is deliberately encouraged for sidewall passivation. Regular removal of etch by-products (that help form the passivation layer) from the reactor walls is necessary to avoid particulate events.
 Figure 4. Plasma ion flux vs. gas flow rates. |
Most semiconductor fabs use two types of chamber clean, a "dry clean" (plasma) and a "wet clean." Dry cleans employ chemistry such as NF3 and Cl2 to etch wall polymers. Wet cleans involve a complete chamber strip to chemically clean parts and can result in tool downtime of up to six hours. Ideally, dry cleans would predominate, since they allow for greatly reduced tool downtime. Optimization of the dry-clean recipe is key to reducing wet-clean downtime. Although wet cleans are still necessary, an increase in the potency and effectiveness of the dry clean can push the MTBC of the wet cleans further out. The greater the time between these wet cleans, the greater the increase in tool uptime and OEE.
 Figure 5. Plasma ion flux measured during the plasma clean over time for four different products. Following the trends in ion flux allows optimization of the clean cycles. |
Altis Semiconductor has optimized the dry-clean recipe by maximizing the measured ion flux. The original dry-clean recipe consisted of an NF3/Cl2/O2/He mix, with each gas flow adjustable to optimize the clean. Other process inputs that affect clean recipe potency are the RF power and the process pressure. The standard approach to optimizing such a recipe involves running a design of experiments procedure, measuring the corresponding effects while changing all the process inputs. Such a process is particularly laborious in this case, since the measured response is difficult to gauge.
Driven by the need to improve tool uptime and reduce downtime spent on wet cleans, and working with varied process inputs, measurement of plasma ion flux in real time was the approach chosen for improving dry-clean efficiency.
Figures 3 and 4 show the response of the measured ion flux to dry-clean process inputs. Although plasma principles can predict the basic ion flux tendencies, it is very difficult to estimate at which pressure or gas flow the ion flux will saturate. It is also difficult to quantify the impact of a particular gas on the ion flux for a given reactor and process. Using the ion flux probe, engineers obtained these results directly, avoiding the time-consuming process of trial and error often involved in tweaking each secondary parameter.
Altis Semiconductor studied the ion flux, changed all parameters at once with full knowledge of the impact of each, and achieved success in hours instead of weeks. Engineers adjusted the dry-clean recipe according to the measured response of the ion flux to each process input, and then ran plasma dry cleans regularly with the new recipe. They were able to eliminate many of the wet cleans that they would normally perform and increased the MTBC by a factor of four.
Predicting chamber clean requirements
The ability to predict the need for chamber clean, instead of cleaning based on RF hours accumulated, increases plasma process efficiency. This ability may be particularly useful to fabs that run many different products and processes. Rather than calculating MTBC for a range of processes, engineers could save time and other resources by accurately predicting chamber clean cycles.
The need for chamber clean depends on the wall conditions. Logically, if the ion flux could accurately determine the state of the wall, it could determine process conditions in real time and predict when to clean. Working on this basis, Altis Semiconductor is now investigating the possibility of using the ion flux probe as an in-line monitor to determine when wall conditions deteriorate.
Figure 5 shows the ion flux during dry clean as a function of lot number for the lots run after a wet-clean process. (Measuring the ion flux during the plasma-clean step provides an indication of the amount of contamination on the chamber wall before the next lot of wafers is processed.) Through 16 lots, engineers ran a dry clean at the beginning of each production lot. The flux values on the y-axis are negative because the ion flux signal is inverted, due to a filter on the signal output. The higher the ion flux signal, the more negative the value.
At the beginning of the wet-clean cycle at lot #1, the ion flux was optimized and the plasma clean carried out. Then, a full lot of production wafers was processed, resulting in contamination on the chamber wall. This action affected the plasma in such a way that the ion flux was reduced. Before the second lot of wafers was processed, a plasma was switched on to clean the contamination from the chamber wall. The ion flux decreased, since the plasma efficiency was reduced due to the chamber wall contamination. As each wafer lot is processed, a reduction will occur in the ion flux as the walls become more and more contaminated. The plasma clean will help reduce the amount of contamination on the chamber walls.
Between the 17th and 18th lots, 36 wafers were processed without running a plasma clean. This resulted in a large decrease in the ion flux, which is evident after the 18th lot, due to large buildup of material on the chamber wall.
The aim of the work at Altis is to monitor the reduction of the ion flux signal and use this to predict when a wet clean is necessary. Optimizing the dry-clean process will increase the ion flux levels and hence reduce the buildup of material on the chamber wall. Monitoring the trends will allow Altis Semiconductor to further optimize its dry clean, and to implement statistical process control limits to trigger chamber cleans.
Conclusion
Reliance on nonproduct wafers for ex situ process control and machine-state parameters for in situ process control is insufficient to meet future manufacturing needs. In situ plasma process monitors can be used for effective plasma process control. By also providing real-time feedback on process anomalies, the ion flux probe allows engineers to measure process repeatability in situ. As a result, end users can maintain uptime throughout the manufacturing process and significantly reduce nonscheduled maintenance.
Roisin Cheshire earned her BS and PhD from Queen's University in Belfast, Northern Ireland. She worked as an applications engineer and product manager for Oriel Instruments, and as an applications engineer for Optronics Ireland in Dublin. She is currently the European sales manager for Scientific Systems in Dublin, Ireland.
John Scanlan earned his BS and MS from University College Dublin, and his PhD from Dublin City University. He was a member of Applied Materials' technical staff in Santa Clara, CA, and account technologist for the company in Edinburgh, Scotland. He is the technology manager for Scientific Systems, 111 North Market Street, Suite 621, San Jose, CA 95113; ph 408/995-5975, fax 408/351-3623, e-mail [email protected].
Gary Skinner obtained a Higher National Certificate in electrical and electronic engineering from North Wirral Tech in Birkenhead, UK. He has been an etch process engineer and service engineer at Applied Materials Inc., in France and the UK, and has worked in etch technical support and as an etch equipment service engineer for Electrotech Ltd. (now Trikon Ltd.) in the UK. He has also served in the Royal Air Force. He is currently a senior etch engineer at Altis Semiconductor in Corbeil-Essonnes, France.