In recent years, pulsed RF plasma processes have been shown to improve etch selectivity, etch anisotropy and reduce feature damage in the semiconductor industry. Carefully selecting the pulse frequency and duty cycle allows for plasma chemistry selectivity and ion impact energy control, both of which are critical for the 7 nm node and beyond. RF carrier frequencies typically range from 400 kHz to 60 MHz, modulated with pulse frequencies in the range of 100 Hz – 50 kHz. Duty cycles range from 90% down to a few percent, and the RF carrier frequency is sometimes adjusted for fast tuning. These together pose a unique diagnostics challenge for troubleshooting when wafer defects occur. For this, Impedans developed the Alfven 100, and this case study explores the pulse monitoring application designed to meet the needs of the semiconductor industry.
The acceptable process windows for 7 nm and below are narrow, particularly for etch rates. Since the duty cycle controls the ion energies, small deviations in the average duty cycle over a process will result in a shift in ion energies and thus etch rate. This drift in the pulse on-time can be caused by many factors – misshapen pulses due to cable degradation, generator pulsing issues and generator calibration errors. Troubleshooting usually requires directional couplers and an oscilloscope, which are expensive, slow and inconvenient, and data analysis is labour-intensive. Even with all this equipment, pulse issues are intermittent, and recording every single pulse with an oscilloscope is not practical.
The Alfven was integrated seamlessly into the 50 Ohm cable and monitored every single pulse with 1 microsecond resolution. Pulse frequency and duty cycle were compared with expected pulse parameters. Atypical pulses counted and reported to the Fault Detection and Control (FDC) system. A predetermined number of ‘bad’ pulses per process were allowed before action was taken.
Impedans and the process engineers verified and compensated for the low-impact of the Alfven unit on the pulse waveform and power delivered to the match box through an adjustment of the 50 Ohm cable length. Then the Alfven was integrated with the FDC system using the simple Ethernet communication protocols available on the sensor.
Fig 1: Once-off verification process to confirm the Alfven has no process impact
Data was accumulated with products being inspected in the usual manner. The number of ‘bad’ pulses was then correlated with the health of the product to determine how many could occur without adverse effects. Similarly, deviations in the average duty cycle for each process step were correlated with wafer defects for a process alarm.