Improving CD-AFM measurements from the tip down

(Nanowerk News) As the sizes of computer chips in electronic devices continue to shrink, traditional measurement tools (e.g., microscopes utilizing visible light) are no longer capable of examining surface features, which can be just tens of nanometers. In order to examine these tiny features, researchers use tools such as the critical dimension atomic force microscope (CD-AFM), which measures features by dragging its ∼10 nm tip across the subject’s surface much like a record player needle pulling across a record’s grooves. With its extremely high accuracy—potential width uncertainty as low as 1 nm—the CD-AFM is the tool of choice for numerous scientists in the semiconductor industry. However, accurate measurements using the CD-AFM are entirely dependent on whether the instrument’s tip is properly calibrated. To address this critical need, PML researchers, led by the Engineering Physics Division’s Ron Dixson, have created reference samples that allow tip calibration with an uncertainty of less than 1 nm.
 Scanning electron microscope image of a SCCDRM sample, displaying the pattern of near-vertical sidewalls with varying linewidths, high uniformity, and known linewidth values
Figure 1. Scanning electron microscope image of the SCCDRM sample, displaying the pattern of near-vertical sidewalls with varying linewidths, high uniformity, and known linewidth values.
“With an AFM, linewidth calibration and tip-width calibration are basically synonymous,” Dixson says. “A linewidth measurement with an AFM depends entirely on the tip calibration.”
When examining the raw data from a CD-AFM, The apparent linewidths actually include the width of the tip. The tip width is then subtracted from the image to provide the final data. If the user believes the tip width is 10 nm but it is actually 15 nm, the uncertainty of the tool’s measurements can rise to an unacceptable level.
So, how do you know the precise measurement of a nanometer-sized tip when the tool in question is one of the only ways of measuring a feature that small?
Dixson and his colleagues have been working periodically over the past 10 to 15 years on a solution to calibrate CD-AFM tips. The first NIST standard for this calibration was created by the team back in 2005 ("Traceable calibration of critical-dimension atomic force microscope linewidth measurements with nanometer uncertainty"), and it provided a tip width uncertainty of approximately 1 nm. This was the result of a fruitful collaboration between NIST and SEMATECH that sent Dixson on assignment at SEMATECH for three years starting in late 2001.
“We didn’t have a CD-AFM at NIST,” Dixson explains. “We had no real AFM linewidth measurement capability, and we weren’t actually expecting that we would get it. Those instruments cost over a million dollars at the time. SEMATECH had an AFM, and they didn’t actually have anyone there to run it at the time. We had AFM experts at NIST, and they had an AFM.” Thus, the collaboration was born.
During this initial research period, Dixson and his colleagues used the CD-AFM and a transmission electron microscope (TEM) in concert to calibrate the samples. This was done by using CD-AFM as a comparator between the samples sacrificed for TEM and those distributed to end users. Although the TEM is a destructive measurement technique, it creates images that allow the researcher to count the atomic planes of the sample.* The NIST researchers were able to measure samples with the non-destructive CD-AFM and anchor the results to the SI meter using measurements with high resolution TEM.
The final samples were distributed to SEMATECH member companies. The end product was a 1 cm2 piece of silicon with multiple measurement targets sunk into the center of a 200 mm wafer. Prior to this NIST effort, researchers used a variety of commercial products to attempt to calibrate their AFM tips. Some of the solutions had a tendency to damage the instrument tips. If the tip survived the process, the products were only able to offer an uncertainty of approximately 5 nm. Other industry scientists would use production samples with known values to attempt tip calibration, again achieving uncertainty of approximately 5 nm. The samples fabricated by the NIST team filled a critical need in the industry, where acceptable standards for calibration of AFM tip width were unavailable.
“The uncertainty was roughly 1 nm,” Dixson says. “The description is approximate because there were only a few chips from that generation that had that precise uncertainty. The majority of the samples had uncertainties closer to 1.5 or 2 nm.”** The difference between samples was due to the variation in linewidth along the features, which were not quite uniform.
Recently, after revisiting the sample fabrication process, the NIST team improved upon the results, creating reference samples that are more consistent from chip to chip and can provide an uncertainty of less than 1 nm ("Process optimization for lattice-selective wet etching of crystalline silicon structures").
 Scanning electron microscope image of a CDR50 type CD-AFM tip
Figure 2. Scanning electron microscope image of a CDR50 type CD-AFM tip. These tips have a nominal geometric width of 50 nm, but the width must be calibrated for each tip.
“The improvement of the new samples is up to a factor of five, taking into account the non-uniformity of earlier ones versus the new ones using the better etching process.”
Like the sample back in 2005, the new reference sample is a single crystal dimension reference material (SCCDRM), which was fabricated using a lattice-plane-selective etch on (110) silicon. Essentially, a pattern of near-vertical sidewalls with varying linewidths, high uniformity, and known values was etched onto a chip (see Figure 1). A user can calibrate a CD-AFM tip by measuring the sample, comparing it to the known value, and making adjustments as necessary.
“In most AFMs, the user has the capability of inputting the known tip width into the tool,” Dixson explains. If the user enters a value and then measures the sample, the results would indicate whether the known tip width is actually accurate.
The improvement of this latest batch of samples can be traced back to the use of potassium hydroxide (KOH) as the etchant instead of the previously used tetramethyl ammonium hydroxide. During the fabrication process, an oxide hard mask is applied to the chip surface. Before the sample can be completed, as much of the hard mask as possible needs to be removed. The KOH etchant was extremely successful at removing the oxide mask without damaging the narrow sidewalls, maintaining the precise features of the structure. On the 2005 samples, the dimensions of the sidewalls were not uniform throughout their lengths, which made user error a possibility if the tip was not placed at a precisely defined spot along the sidewall.
Since this research was completed, Dixson participated in a comparative study with PTB (Physikalisch-Technische Bundesanstalt, the National Metrology Institute of Germany). PTB was also working on a similar calibration solution for CD-AFMs, and it made sense for the two institutes to cross-check their samples. The soon-to-be-published results indicated that the samples from the two institutes had very similar levels of uncertainty.
Samples from the NIST team’s new and improved effort have not been widely distributed at this time. A wider distribution is under consideration. In addition, a future focus on providing additional CD-AFM tip calibrations is being discussed.
“We might explore the possibility of calibrating the vertical edge height,” Dixson says. “That is the distance between the base of the tip and the point that contacts the sample. There is no official standard available for this.”
*Strictly speaking, the HRTEM image does not allow the researcher to count the atoms. The researcher is actually counting interference fringes. But, there is a one-to-one correspondence with the atomic planes.
**The NIST master sample from the project is known to 0.6 nm (k = 2), but the added uncertainty due to the transfer process meant that the distributed samples all had larger uncertainties.
Source: NIST