Defect Control - Designing for Process Cleanliness

By Dr. David Kirkwood, Senior Project Manager, Contamination Control,
Axcelis Technologies, Inc.

The defect performance of capital equipment utilized in semiconductor manufacturing is a critical element in determining the yield and performance of the devices thus produced. Yield is directly impacted by both the number & size of those defects commonly referred as particulate adders, which are inadvertently added to the wafer as a byproduct of each process module. Furthermore, fab productivity is impacted by “particle downs”, whereby particle excursions of sufficient magnitude above the UCL may result in one or more tools being taken out of production in order for the problem to be addressed. Hence it is of critical importance that all manufacturing equipment be designed to minimize any sources of particles or other contaminants.

Evolving Trends
As device geometries have scaled progressively smaller with each generation, so have the requirements on the permitted levels of particulate contamination (and other associated defects, such as metals – see ITRS 2011 roadmap [1]). To understand this evolving scenario, let us consider a typical particle spec for a device manufacturer in 2013:

  • <30 adders @ > 0.045 µm with 90% success rate
  • 0 defects at @ > 0.2 µm

The primary specification is tied into the yield calculations which govern operational productivity. If the implanter is able to run process modules with this particulate level, then the yield will not be impacted. In practice, process modules are not equally sensitive to particulates because it is intimately tied in to the step which follows.

Above a certain size, particles may become “killer defects”, in that they will either prevent the implant step itself from being finished correctly (through shadowing of the doped area or otherwise), or resulting in a physical change to the local morphology that will effect subsequent deposition steps. As device geometries shrink, by approximately a factor of 0.7 per generation, so too will the minimum critical dimensions of any surface defects which results in device performance degradation. Hence the specifications for capital equipment typically reduce the lowest bin size by a similar factor for each device generation.

In addition, the absolute number of particle adders is expected to be lower for each generation. The reasons for this lie in the growing complexity of device fabrication; as the total number of process modules increases, the added defect level per module must fall. Concomitantly, increasing pressure on margins result in the removal (as far as possible) the number of cleaning stages between value-adding process steps, hence tolerance to added defects is reduced.

The picture is further confounded by ongoing requirements for simultaneous compliance. Historically, each specification in a technical acceptance document was handled in isolation. Increasingly however, especially in the Foundry environment, compliance to throughput, uptime and defect levels are required to be simultaneous.  Some Logic manufacturers have sacrificed ultimate throughput in order to attain the required success rate at a given particle spec, however other device manufacturers have been slow to follow this paradigm.

Particle Generation in Ion Implantation
Ion implanters can produce particle contamination by a number of different primary mechanisms. In particular, the following are unavoidable bi-products of extracting and mass-analyzing ion beams: direct sputtering (or thermal damage) to implanter vacuum surfaces by analyzed ions or neutrals; delamination of deposited coatings from unionized process gas or photoresist residue; electrical discharge between high voltage electrodes and grounded elements in the beamline. Secondary sources, which arise from non-value adding activities such as preventative maintenance, include residue from in situ cleaning activities; human contamination; vacuum leaks.

A key feature of particles generated in ion implantation is that they do not follow a normal distribution. This applies both to the number of defects per wafer, and the bin size distribution. Typically the former is a lognormal or similar distribution (for example, see figure 2 (NB – herein 0 adders are treated as 0.1, which results in a skew in the distribution)), wherein most wafers processed will have low adder counts, however a few may have high excursions. This must be taken into account when constructing a particle spec. Upper control limits based around a mean and standard deviation are valid for normal distributions of data – for lognormal, the mean is not a good metric for control, however the number of excursions above a limit are.

Typical Particle Distribution

Figure 1: Typical Particle Distribution

The particle distribution as a function of bin size is also non-linear. This is a consequence of both particle generation (ion beam interaction with graphite liners being one of the primary source in ion implantation), and transport mechanism whereby small particles are more readily entrained within the beam sheath unless otherwise prevented by a potential barrier. When one takes into consideration both of these elements, reduction in the absolute number of adders, coupled with a lower limit on the critical dimension measured, represents a fundamental process challenge.

All ion implanters, irrespective of design or manufacturer, can and will suffer from any or all of the above. The key is to design the machine such that opportunities for the secondary sources are eliminated as far as possible, and that the former unavoidable sources are mitigated through equipment design and process control. Particle issues exhibit a high degree of process specificity, hence it is non-trivial to completely design out opportunities for defect mitigation in new tools.

Design for Process Cleanliness

In order to control particle excursions in the implant process, it is vital to control two key elements associated with these defects – generation and transport. The ion implanter must be designed to minimize opportunities for generation, and to interrupt transport of the particles. Some examples of this are detailed below.

The primary means of mitigating unavoidable particle generation through ion beamlets interacting with vacuum liners is through prevention of misalignment of the extracted ion beam. This is of particular importance in the extraction optical region near the ion source, but is also true throughout the entire beamline of the tool. Minimizing thermal drift, which thereby reduces misalignment over time, eliminates one of the major sources of particle elevation, particularly at extended source life. One of the fundamental limits on PM frequency of the source region is that the particle levels can become elevated above a certain number of source hours – the Eterna ELS3 source technology deployed by Axcelis routinely attains >400 hours in production within spec particle levels.

Once the location of the ions is tightly controlled, the next step is to optimize the material and geometry of those components which will see beamstrike. This includes choosing materials which have good thermal stability and angling these to minimize sputter yield. Furthermore, the morphology of liners in areas adjacent to beam transport is tailored to retain deposited material and prevent flakes from falling into the beam and generating secondary particles. Working closely with our suppliers, Axcelis has qualified a bespoke grade of vitrified graphite on all components subject to high levels of electrical stress, providing optimal particle performance whilst maintaining low CoO.

Controlling deposition along the beamline elements is also critical to preventing a long term drift in the particle baseline. Deposited materials arise principally in part to the short residence time of process gas in the arc chamber, and in part from outgassing from the wafer surface. To this end, elements of the beamline remote from the beam path are cooled to provide preferential deposition zones which will not interact with the ion beam, and improve vacuum conductance in critical areas of the tool, such as the extraction region, to maximize the degree to which unwanted gas is pumped from the system.

As a consequence of the voltages required to extract and control an ion beam in the implanter, High Voltage (HV) stability is key to maintaining particle levels. One high voltage arc can result wafer scrapping due to the number of particles added. To this end, Axcelis tools are all fitted with proprietary HV shields, which prevent arcs from establishing in the event of high voltage breakdown events, thus preventing catastrophic yield loss. In conjunction with this, a specialized grade of graphite is used for these electrodes, which provide uniform electric fields and minimize surface residue (the most common reason for arcs being instigated).

In the production environment, extended downtime following scheduled preventative maintenance, may occur as a result of in situ cleaning / alignment. The Design for Serviceability methodology has been evoked throughout the design of the Purion platform tools, and these include the removal of all in situ cleaning by having major modules readily removable from the tool, and used with vacuum chamber liners which are swapped out during the PM. These elements both remove opportunities for defect introduction, and reduce the overall time of the PM, thus reducing downtime (simultaneous compliance to the requirements of the customer).

Finally, in an ion implanter most particles are transported the wafer through entrainment in the sheath of the ion beam. The AEF module in the Purion M implanters acts as an effective potential barrier preventing particle propagation to the wafer.

The Purion platform of ion implanters have been, from their inception, designed with contamination control as a primary focus, using a protocol named “Design for Process Cleanliness (DfPC). This is an integrated, holistic approach generating a number overarching design principals, some examples of which were detailed above for the beamline of the tools. Analogous design criteria have also been incorporated into the common end station of the platform, including the elimination of vertical wafer handling and optimizing loadlock operation to prevent pressure bursts / condensation sources of particulates.

In figure 2 below, typical in production particle performance is presented for the Purion M implanter, demonstrating the performance levels expected as a result of DfPC.

Purion M Production particle data

Figure 2: Purion M Production

Concluding Remarks
Particle requirements will continue to challenge the industry going forward, and will require ongoing investment in both metrology and design effort. Through partnership with key customers, we can work to minimize the costs associated with this work, whilst at the same time delivering solutions which attain the required defect performance for current and future device generations.