Power Supplies Advance Beyond Volts and Amps

Everyone in the semiconductor industry is in the business of developing systems that provide benefit to the user — whether it is the consumer, a device maker, chip developer/producer, foundry, an OEM or a component manufacturer. It is appropriate that this effort should use available engineering achievements designed specifically to increase the ultimate goal's effectiveness, namely cost efficiency without sacrifice.

IC processing systems commonly use power supplies to regulate certain parameters (power, voltage or current), establish the set point, and turn the unit on or off. Sometimes, the sophistication goes as far as to read back the unit outputs, maybe using these parameters for some system-level process checks. However, through the years, power supply manufacturers have learned more about customer processes and have added many features that greatly enhance system operation. Unfortunately, users often fail to take advantage of these features.

The importance of accessing a power supply's varied features is obvious. Accurate, repeatable delivery and transparent process variation handling are two immediate issues facing system design and process engineers. These goals, coupled with power supply technology in existence, are summarized in Figure 1. Power supply features can add value to any process. This article highlights many of the features developed within power supplies, along with suggestions on how to implement them to improve cost efficiency without sacrificing performance or reliability.

One feature is the use of communications. There are two basic conventions available: serial (host) and analog (user) communication. Typical serial communications include Ethernet, DeviceNet, PROFIBUS, RS232, RS422, RS485, an active front panel or a remote control panel. Each has advantages, such as convenience, driver support, availability of utilities, cabling reduction and sophistication. All have improved set-point granularity and easily accessible, full-featured command sets. Also, the connection itself is made easier and more robust with standard, off-the-shelf designs that produce obvious faults given an intermittent connection.

1. Significant benefits arise from using advanced power supply features available in today’s systems.

Serial communications are ideal for setting up a particular process, but may not be sufficient for time-sensitive information like turning the power supply on and off, or gathering real-time process parameter information.

Conversely, analog communications can be very rapid, and the information can be transmitted via several parallel paths (for example, a 37-pin user port has many paths). If a unit is to be turned on or off, or if the system needs to retrieve the process voltage at a particular point in time (endpoint detection), there is no infighting between which command has priority — both are processed simultaneously. The information is instantly transmitted or received. The obvious disadvantage of the user port is that it has limited access to the power supply's full capabilities. A skilled operator familiar with the unit's functionality may be required for system setup if the default parameters are not optimized for the specific process.

If the system is designed properly, the best of both worlds often can be achieved: process initialization with a serial interface and process execution/real-time data gathering with the analog interface (Fig. 2).

2. A combination of interfaces maximizes system performance, yielding better process control.

Processes are rife with variations: problems encountered during ignition; ramping power to a required set point; accounting for the occasional arc and the way the power supply handles it; changes in gas delivery, which causes the chamber impedance to vary — the list is long. Some systems try to account for these when determining an endpoint. Most systems, however, do not compensate at all but use a simple timer to turn the power supply on and off, which provides only a crude estimate of how much deposition has actually occurred.

A significant advancement is the capability to measure power delivery directly from the power supply instead of making process assumptions at the system level. With this feature, special counters tally up to a user-specified number of joules, and internal circuits cause the power supply to turn off once the counter reaches that predetermined value. The joule mode counter measures delivered energy and increments the counter every 2 msec (1 msec with newer product generations). This way, the unit compensates for process variations and delivers accurate quantities of energy during each run, regardless of variations in the number or length of arcs, or in process gas pressures, etc.

Figure 3 shows a simple sputter process of 18,000 W for 1 sec (18,000 J) that includes a 20 msec delay with ignition and 17 msec ramp, and some arcing and impedance variation. At the end of 1 sec, there was nearly a 7% error that would not have been accounted for with a standard, 1 sec timer. Use of the joule mode compensates for this by enabling the unit to run an additional 70 msec, bringing the run-time error down to a negligible value. With the joule mode compensating for run-to-run variations, the actual run time would vary from one wafer to the next and, with this feature enabled, the resultant run-to-run error becomes very small.

3. Deposition may not be accurate from run to run. In this example, a 1 sec process results in 7% error (assuming constant process power). Joule mode would compensate by adding 70 msec of sputter time.

Designers are developing systems with a competitive advantage that includes a robust multiple process material capability. Different materials sputter at different impedances and, as such, arc behavior changes as well. Simple aluminum targets operate at ~600 Vdc, and their arcs are distinctive because the impedance drop is well defined. Metals such as silver operate at much higher impedances (closer to 1000 Vdc), and an arc impedance drop would not be the same. Similarly, how quickly an arc extinguishes may also be material-dependent. Having the ability to tailor the system to meet arc handling needs from different materials is critical to leveraging a configurable system.

Arc handling variability and programmability are standard within current power supplies. Configurable values include voltage arc trip level and arc delay time, which are used to define an arc's existence. Another is arc shutdown time, which can vary the recovery time once the handling algorithm has been invoked.

The voltage arc trip level and the arc delay time define whether an arc exists. If the process voltage falls below the voltage arc trip level for a longer duration than the arc delay time, the power supply classifies the event as an arc. When the conditions are satisfied, the energy source is removed from the circuit, limiting damage. Because significant process benefits can result from the power supply's arc handling method, this must be researched and studied carefully before investing in a power supply. The vendor's applications engineers can help in choosing the values that best suit a process.

Once the arc is detected and the energy is interrupted, power is not restored until the arc shutdown time has expired. This time can be shortened to speed up recovery. Or, if the arc is caused by a thermal hot spot, it can be lengthened, which allows the arc zone to cool.

Figure 4 shows how a power supply can recover from an arc. By changing the three variable arc handling parameters — accessible via the serial interface — a system can customize its response based on the types of materials being used. A system's architecture can be designed so that, as a target changes, the user is prompted to specify the material. Modified arc handling parameters can then be downloaded into the power supply for optimum performance.

4. In the events taking place while handling an arc, the voltage arc trip level is 150 Vdc, the arc delay time is 0 µsec, and the arc shutdown time is 20 µsec.

Determining sputtering process efficiency is crucial. Wafer yield measurements are critical to understanding overall system efficiency. A way to measure quality is by understanding the number of arcs that occur during a process run. This is a readily accessible parameter, enabling the system to make a gross determination.

Two methods exist to help determine this parameter. The first is reading the arc count from the supply directly after a run (via the serial interface). Comparing the arc count with a threshold value allows a first go/no-go quality determination. If this is flagged as an issue by the system controller, the system also knows where a correction may be investigated.

The second method of arc counting involves a comparison within the power supply. The system initializes an arc count threshold via the serial interface. If the resultant arc count for a run exceeds this threshold, it will be indicated by a digital (high/low) signal on the analog user port. Simply monitoring this signal on the user port after every run can provide an input to a system-quality algorithm. The system can act on the signal immediately by marking the product as bad, or it can trigger a subroutine that queries the supply for additional information such as actual arc count. This subroutine can then help determine whether the product has sustained irreparable damage.

Any process can "run away." Therefore, if it is possible to help prevent any damage from occurring and to identify this state when it takes place, it becomes possible to get closer to system optimization.

A simple way to use the power supply to prevent damage is to program the unit to remain within certain system-configurable power, voltage and/or current thresholds. Three commands within the repertoire of the serial (host) command set enable these functions. When the programmed limit is set and a runaway condition causes it to be reached, the unit will not be able to achieve set point. By monitoring the set-point signal on the analog (user) port — along with the power, voltage and/or current analog pins — a system can quickly detect what is happening, classify the situation and presumably correct the problem. This is a proactive method for determining when a problem exists and, subsequently, preventing damage with a subsystem that can react quickly to rapid process changes.

Target use is key to reducing system material costs and maintenance time, and to improving overall throughput. The difficulty with trying to squeeze everything out of a target is that erosion patterns cause impedance changes that may vary the process (i.e., the incidence angle of the arriving ions will change). This reduces sputtering efficiency, and more power may be required to get the same throughput.

The target life counter uses energy accumulation methods similar to the joule mode to determine total energy delivered during the target's lifetime (that is, since the system counter was reset). The system controller, through the serial interface, can send the overall target life (in kilowatt-hours) to the power supply. As energy is delivered to the target, the power supply reduces its internal target life counter, so that its value represents the target's remaining life. A system subroutine can then read back the remaining life — also via the serial interface — and apply a system-established compensation factor to such variables as system power and/or gas flow control.

Because of the multitude of targets in use today, a statistically significant number of samples should be run to identify actual lifetime and compensation factors.

When used correctly, an efficient target conditioning cycle (TCC) can improve system uptime by providing an efficient, rapid method for removing the surface oxides of a newly installed target, ensuring the first products processed are not damaged by residual material.

Classical target conditioning requires significant firmware design work and a lot of processing power. A way around this is to maximize the available power for a short period of time; however, this can lead to significant particle damage on the first products processed.

Recent power supply design advancements optimize TCC capability by detecting the correct power levels required to properly condition a target. This TCC feature works with a sensor that determines when significant arcing occurs. The algorithm then backs off the power level until the arcing subsides, then seeds itself with a new power set point and begins ramping up. If the arcing has subsided, it continues to increase power in this manner until the unit reaches the original set point. Having the power supply take over this function simplifies system development while ensuring a consistent method of proper target conditioning.

Adding a reactive gas to the sputtering process complicates matters since the resultant (and desired) material is often an efficient insulator. This dielectric material can then coat the target, leading to a charge buildup and heavy arcing.

Past processes addressed this phenomenon by using a high-frequency ac supply (i.e., 13.56 MHz) or a complicated bias scheme that alternately neutralizes the destructive charges. These solutions are expensive, require additional components to assist in matching the design impedance of the power delivery unit, and can lead to excessive substrate heating. Even after negotiating these system obstacles, when sputtering with high frequency the user is left with a relatively low sputter rate, reducing system use.

In the early 1990s, the ability to neutralize this charge buildup by periodically reversing a dc supply's voltage was introduced. Throughput improvements approaching dc sputtering rates were realized while maintaining system simplicity. Soon, exotic oxides and nitrides were becoming standard materials in optical thin-film and data storage applications.

Additional "knobs" have been developed with regard to process optimization, including frequency and reverse time adjustments of the pulsed dc waveform. Varying these parameters affects the thin film's quality. Hardness, clarity, granularity and film stress are the new metrics describing these materials, and pulsed dc power enables the new horizon.

Just how much reactive gas gets into the chamber can determine not only process efficiency but can also regulate the dielectric film's correct chemical makeup. Operators have been forced to contaminate their systems with excessive reactive gas amounts to achieve the minimum results. However, by doing so they compromise throughput because the sputter rate suffers. Reducing the gas too much brings the process back into a higher rate, but an unacceptable metal mode.

System optimization while maintaining the stoichiometry involves judicious use of the reactive gas. It is now understood that measurable parameters exist, and that the reactive gas's perfect balance with metal sputter rate can be maintained. With a closed-loop control system that quickly measures the parameters and adjusts the partial pressure, the system can cost-effectively produce the new materials.

Only a few power supply features that can benefit end users can be summarized here. Many within the data storage, optical coating and flat-panel display industries have been early adopters of these features, and have worked to refine them. Some of these features are easily implemented whereas others, like the investment in and use of a serial interface, may require system reconfiguration. Implementation cost must be weighed against improved performance and/or yield to determine whether the cost is justified.