Roland A. - PVD for microelectronics (779636), страница 33
Текст из файла (страница 33)
Instead, mechanical or electrostatic clamping is required. Another concern is the robotic arm itself, which can be a source ofunwanted particles and contamination. Unlubricated moving surfaces inSPUTTERING TOOLS169contact (e.g., the bearing surfaces in the arm) can generate fine particlesthat reduce device yield. Unfortunately, wet oil-based lubricants tend tooutgas and create molecular contamination, while dry lubricants can createas many or more particles as the bare contacting surfaces themselves.
Also,an arm is often required to hand off or pick up a wafer in a process chamber at elevated temperature ( > 400~and the radiant heating can causegrease-related outgassing.With regard to wafer holding, it is useful to consider how rapidly awafer clamped to a horizontal wafer platen can be accelerated or decelerated before it begins to slip since this has a strong influence on howrapidly the wafer can be moved from one stationary position in the PVDtool to another.
Elementary mechanics shows this acceleration is a =txF/M, w h e r e / x is the coefficient of static friction between the wafer andplaten (e.g.,/z ~ 0.3 between Si and A1203), M is the wafer mass ( ~ 50 gmfor a 200-mm Si wafer), and F is the total vertical clamping force: clamping pressure x clamping area. For a wafer mechanically clamped at itsedge, the maximum tangential acceleration is typically ~ 2-3 g (1 g = theacceleration of gravity at the earth's surface = 980 cm/sec2). The full-faceholding of an electrostatic chuck (ESC) leads to a much greater clampingarea nearly as large as the wafer itself, leading to perhaps a tenfold increase in maximum tangential acceleration.
Therefore, while ESCs aregenerally thought of as a clampless way of holding wafers stationary during processing, they are compatible with rapid handling of wafers betweenprocess steps.The typical application for a vacuum handler in a radial cluster tool is totransfer wafers between different process modules that are themselves vacuum-isolated from the transfer module by a slit valve. In view of the angular rotation and linear translation needed to effect this transfer, the simplest handler requires three rotating points: shoulder, elbow, and wrist.
Arepresentative handler of the "frog-leg" design is shown in Fig. 5.41,where a dual robotic arm has been incorporated to handle two wafers simultaneously for improved throughput. Wafer transfer from one module toanother involves at least five separate motions (e.g., linear motion into andout of module 1; rotation to module 2 position; motion into and out ofmodule 2). Since advanced PVD process sequences require multiple modules (3-5 or more), reliability of robotic arms is of great concern. It hasbeen estimated that a mean-time-between-failure (MTBF) of > 106 cyclesis required to avoid impacting overall PVD cluster tool performance.Rotary motion is a particular challenge for a vacuum robot since this requires coupling of the arm, which is under high vacuum, to the motor,which is out of the chamber at ambient pressure (760 Torr).
Vacuum-tight170R. POWELL AND S. M. ROSSNAGELFIG. 5.41 Representative vacuum robotic handler of the "frog-leg" design (courtesy of BrooksAutomation, Lowell, MA).sealing of the rotating-shaft connecting arm and motor is often accomplished with a Ferrofluidic TM seal, in which a concentrated magnetic fieldis used to retain a ferrofluid (ferrite particles suspended in a low vaporpressure fluid) in an annular gap between the shaft and the magnetic components surrounding it (see Fig. 5.42). Direct-coupled rotary feedthroughswith Ferrofluidic seals allow rotary operation in vacuum at high speed andhigh torque. Another approach to rotary motion is to indirectly link themotor and arm by means of magnetic coupling.
For example, a permanentinternal magnet fixed to the shaft can be used to track the motion of a rotating external magnet in air (see Fig. 5.43). The simplicity of the linkageis offset to some extent by limited torque transmission, backlash, and thedifficulty of coupling at high rotational speed.A variety of robot designs have been implemented for handling wafersin the high-vacuum ambient of a PVD cluster tool (10-8-10 -9 Torr);however, they all share a common concern with wearing surfaces (e.g.,FIG.
5.42 Schematic of a Ferrofluidic T M seal used to make a vacuum seal to a rotating shaft (courtesy of Ferrofluidics Corp., Nashua, NH)FIG. 5.43 Schematic of a magnetic approach used to couple rotary motion into a vacuum ambient(after Fig. 6 in ref. 5.43).172R. POWELL AND S. M. ROSSNAGELstainless steel or ceramic ball bearings) that shed particles.
As noted earlier, contamination and outgassing generated by surface lubricants complicates the issue. Familiar dry film lubricants with a platelet-type microstructure (e.g., sulfides such as MoS 2 and WS 2) have ultralow vaporpressure even at moderate temperatures (typically < 5 x 10 -12 Torr at20~ and < 5 x 10 -9 Torr at 100~but shed particles at levels comparable to plain dry bearings. While h i g h - v a c u u m lubricants designed forthe stringent particle and contamination requirements of cluster tool processing are a relatively new development, there is growing interest inTeflon-type dry lubrication. One such formulation of note is poly-tetrafluoroethylene (PTFE), which has been incorporated into ball bearing assemblies with a reduction of several orders of magnitude in particle generation rate.5.4 300-mm PVDThe terms "8 inch" and "200 mm" are often used to describe the same waferdiameter as if there were 25 mm in an inch instead of 25.4 mm.
Actually, allwafer diameters since 6 inch have been metric. Therefore, although a 4-inchwafer is in fact 4 inches in diameter, referring to an 8-inch wafer overstatesthe actual diameter by about 1.5% (8 inch ~ 203 mm).When PVD was introduced into microelectronic production in the late1970s, wafer diameter was predominantly 3 inches. By 1997, however, thetotal area of Si used to make ICs ( ~ 4 x 10 9 in 2 per year) was more or lessequally divided between 6-inch (150-ram) wafers and 8-inch (200-mm)ones.
The IC industry has agreed that the next step will be to 300 mm, andthis transition will present a significant technical and economic challengefor PVD I and most other processing as well.The motivation for chip makers in going from 200-mm to 300-mm Si iscost reduction ~ 2.5 times more die can be obtained per wafer. This gainis due to a 2.3 times larger wafer area and a larger edge-to-area ratio thatallows large rectangular die to be more effectively packed on the wafer.Overall, chip makers hope to lower the cost per cm 2 of processed Si by15-40%. We will not provide an in-depth treatment of the hardware implications of processing tools for 300-mm wafers (refs. 5.44-5.48 provideuseful background information on growth, handling, and processing ofSPUTTERING TOOLS173these wafers); however, several comments can be made with regard to thespecific use of 300-mm wafers for PVD.1.
Wafer Cost. When 300-mm wafers of test grade were introducedaround 1993, they cost ~ $1500. Prime device-quality material will bemore costly, and unless production-volume usage greatly reduces 300-mmwafer price, this will be an issue for PVD. In particular, the use of testwafers for equipment development or process qualification will be morelimited, leading to more use of hardware modeling and in-situ metrologyto qualify hardware performance.2. Wafer Dimensions.
Proposed dimensions for a 300-mm Si wafer arediameter = 300 mm ( + 0 . 2 mm) and thickness = 7 7 5 / x m ( + 2 5 / ~ m ) . It isunlikely that nonuniformity of film properties such as thickness, sheet resistance, and step coverage will be relaxed from their current 200-mm levels. Retaining such levels of uniformity (3o- < 5%) over an area 2.3 timesas great will be a major challenge to DC magnetron design. This couldlead to the use of rectangular sources with relative substrate motion similar to what is done when using PVD to coat extremely large-area glasspanels for flat panel display or architectural applications. Also, gas injection and pumping ports for reactive PVD will need to be designed to produce uniform films such as TiN over these larger areas.
Nevertheless, oneexpects PVD processes to scale more easily to 300 mm than do chemistrydominated processes such as CVD and reactive ion etching (RIE).Since the thickness planned for 300-mm wafers is virtually the same asthat currently used for 200-mm ones, the area-to-thickness ratio (zrr2/t)will increase by a factor of 2. Such wafers will be very fragile to handleand particularly susceptible to thermal or mechanical stress. For example,from Eq. (5.4), the central deflection of an edge-clamped wafer for a givenpressure of backside gas and wafer thickness is proportional to r 4, leadingto 16 times more bow in 300-mm wafers.
This strongly argues for the fullface holding and temperature uniformity provided by an ESC. The backside holding of an ESC also avoids the frontside edge exclusion associatedwith a typical mechanical clamp ring ( ~ 6 mm), which for a 300-mmwafer would exclude ~ 8% of the area.3. Cluster Tool. Because many PVD processes are enabled by suchsteps as degas and preclean, it is clear that these processes (and their hardware) must also be scaled up on a 300-mm PVD cluster tool. While scaling up an entire 200-mm tool by (300/200) I/2 ~ 1.2 in all directions is possible but probably not required, the footprint of a PVD cluster tool for 300mm is still expected to exceed that of 200 mm.