Next-Generation Lithography Tools: The Choices Narrow
Ruth DeJule, Associate Editor -- Semiconductor International, 3/1/1999
|
| Designed for all MICRASCAN tools from 248 to 157 nm and EUV, a vertical stage is shown scanning a wafer past the exposure source. (Source: SVG Lithography) |
Eighteen months ago, International SEMATECH narrowed its choice of possible successors to optical lithography to four: extreme UV (EUV), ion projection lithography (IPL), SCALPEL and X-ray. Each has demonstrated printing capabilities to 70 nm or below. However, at the recommendation of the Next Generation Lithography (NGL) Taskforce, funding now will be limited to EUV and SCALPEL. Timing appears to be the factor with 157 nm optical lithography intended to carry the industry through the 90 nm technology node, thus paving the way to an additional decade of development. But with ever-present uncertainties, can we afford to overlook any of the NGL techniques? Proponents of each tell why they continue to develop and promote their respective technologies.
Extreme UV
Building on conventional optical lithography technology and infrastructure, EUV uses 10-14 nm illumination for printing images down to 30 nm (Fig. 1). Low NA optics and good DOF provide a large process window to accommodate isolated and dense lines simultaneously, thus eliminating the need for resolution enhancement techniques such as OPC and phase-shifting. A consortium of semiconductor device manufacturers, including Advanced Micro Devices, Intel and Motorola, called the EUV Limited Liability Co. (LLC), was formed in 1997 to extend basic research and develop an alpha exposure tool.
Operating in a vacuum environment, EUV radiation is produced from a 45 eV plasma by heating a supersonic xenon gas jet with a high-powered laser, typically a Nd:YAG. The radiation is collected by the condenser and shaped into a narrow arc or 'smile' illumination beam and focused on a reflecting reticle. The reflected radiation passes through a 4X reduction camera and is imaged onto the resist covered wafer. The entire patterned field is illuminated by scanning the reticle through the beam. Correspondingly, the wafer is scanned at one-fourth the reticle speed in the opposite direction to reproduce the mask image on the wafer. Conventional silicon processing can be used to define mask patterns on 4X reticles.
|
Fig. 1. Illustrated is an EUV lithography system that has demonstrated imaging 50 nm features with ultra-thin resist layers. (Source: EUV LLC) |
The strength of this optical technique is the potential to accommodate multiple generation technology nodes without loss of throughput, said Charles Gwyn, technology champion from Intel (Santa Clara, Calif.). For a 0.25 NA optical system, he said, adequate process margin is available to scale minimum dimensions while maintaining constant EUV flux at the wafer.
EUV lithography is rallying interest from several stepper manufacturers. For example, a European EUVL research program was recently formed combining the expertise in equipment development/integration of ASM Lithography, optics of Carl Zeiss and synchrotron source technology of Oxford Instruments. The team will focus primarily on potential showstoppers and study alternative sources such as plasma and synchrotrons. Also, SVG Lithography has entered into agreement with EUV LLC to review and critique an engineering test stand and develop a basic beta tool design later this year. According to Gwyn, 2004 is the targeted timeframe for beta testing and early production of EUV systems.
X-ray lithography
Among NGL tools, only one technology offers a pre-production lithography system, X-ray. A decade of experience and development has resulted in the integration of all components, from X-ray source and stepper to resists and defect-free masks. XRL's biggest advantage is that it is known to work, said Jerry Silverman, technology champion from IBM (East Fishkill, N.Y.). Aggressively promoted in Japan, XRL has been used in the fabrication of GaAs devices at Sanders, a Lockheed-Martin company, and to pattern six million transistor, 400 MHz PowerPC microprocessors on 8 in. silicon wafers at IBM (Fig. 2). The 1 nm X-rays mean the absence of thin film interference effects and thus excellent process latitude.
|
|
Fig. 2. Gate-level resist image of a PowerPC 604e MPU exposed with XRL has ;6 million transistors on the chip, operating at ;400 MHz. (Source: IBM) |
Feasibility is not an issue. The most expensive component, the synchrotron X-ray source, has demonstrated reliability >98% and is expected to support 15 or more steppers. Even cost of ownership can be acceptably low for high-volume production, where 12 steppers reduce the source portion of total cost to 4%. The production of 1X masks, however, continues to pose problems in the realm of image placement, CD and defect control and in the development and commercial availability of suitable e-beam mask writers. Can you make a mask and meet all error budgets and yields? 'With a mask area 1/16th that of a 4X mask, we believe so,' Silverman said. Researchers at IBM are currently developing a mask writer with features to meet 100 and 70 nm generations.
X-ray steppers can operate at atmospheric pressure, in a helium environment, use grazing incidence mirrors and instead of a lens system, have a beam line. The current pre-production tools have throughput nearly a quarter of the required production-level 60 wph. X-ray Lithography tools that use synchrotron sources such as those from Canon and SVG Lithography position the wafer vertically to access the horizontal X-ray beam. Canon's pre-production system is installed at Mitsubishi. Later this year, a second-generation production XRL tool, the XRA-1000, will be shipped to ASET for research but is useable in manufacturing.
Point X-ray sources
Generating X-rays for semiconductor lithography applications can be accomplished with GeV superconducting synchotrons or with high-temperature, high-density plasma sources. A synchrotron is economical for high-volume production, because it can feed multiple steppers. However, for process development and moderate volume production, point sources may provide an attractive alternative.
A granular point X-ray source resembles an excimer laser system used in DUV lithography tools. Two approaches are being developed, laser plasma and dense plasma focus (DPF) sources. Science Research Laboratory (SRL, Somerville, Mass.) has a kilowatt-class, DPF source that features an all-solid-state driver with energy 'snubbing' to extend the lifetime of plasma discharge electrodes and improve reliability. Using concentric anode and cathode, kilowatts of usable X-ray power can be generated at the point source for throughput to 20 wph. The DPF source is being integrated with the SAL (South Burlington, Vt.) XRS2000 X-ray stepper and will be used for production of GaAs MMIC chips at Sanders.
The laser plasma X-ray source developed at JMAR Research uses a diode-pumped solid-state laser to produce 150-1000 psec pulses at kilohertz repetition rates. In a helium environment, the source focusses 1 x 1015 watts/cm2 onto a copper target, generating a plasma that spherically radiates 1 nm X-rays. A collimator is used to obtain a synchrotron-like beam. High powers are achieved by combining multiple laser modules in a single system. A beta tool aimed at 24 300 mm wph is currently under development for test and evaluation by 2000.
SCALPEL
|
|
Fig. 3. Using SCALPEL, 80 nm contact holes printed in 7500 Å of DUV ARCH resist had DOF >20 µm for a 10% CD change. (Source: Lucent Technologies) |
Charged particle systems such as IPL and SCALPEL may be comparatively economical, because they do not require expensive optics and costly sources as do EUV and X-ray, respectively. Instead, they use conventional charged particle optical systems and established sources (ion or electron beam). The fundamental issue impacting charged particle systems is beam blur caused by space charge effects. As the beam current is increased, like-charged particles tend to repel each other in a statistically random way (stochastic Coulomb interaction), which causes the image to slightly blur. The maximum available throughput will be largely determined by the maximum allowable beam current. Therefore a decrease in beam density to correct for blur increases exposure times and correspondingly, decreases throughput.
The manufacture of a production-worthy SCattering with Angular Limitation Projection Electron beam Lithography (SCALPEL) tool is currently being discussed as part of a cooperative effort among Lucent Technologies, Applied Materials and ASM Lithography. Using scattering contrast, as opposed to absorption contrast, SCALPEL prints a 4X reduced image of the mask features (Fig. 3). A multilayer membrane mask is designed so that a negligible fraction of electrons is absorbed by the structure, allowing most electrons to pass through. The pattern is formed in a high atomic number layer that acts to scatter electrons as they pass through. An aperture in the back focal plane of the projection system absorbs the electrons from the patterned layer, and a high-contrast aerial image at the wafer plane is formed.
SCALPEL prints linearly and does not require complex resolution enhancement techniques such as OPC or phase-shifting, commonly used for advanced DUV and 193 nm optical lithography. This may translate to far lower mask costs and ultimately lower cost-of-ownership, thus posing an attractive alternative to optical lithography, said Lloyd Harriott, technology champion from Bell Labs, Lucent Technologies (Murray Hill, N.J.).
In development since 1989, feasability studies continue. Because SCALPEL's step-and-scan writing strategy produces stripes that are smaller than the size of a chip, it is necessary to stitch together several stripes per chip. The question is: Can stitching be achieved with sufficient accuracy to meet CD error budgets? While it's still too early to answer, progress is being made to demonstrate various blending techniques to meet feature placement requirements. Similarly, mask and resist technologies continue to advance through joint development efforts. With a commercial supply of mask blanks from MCNC, mask shops such as Photronics and DuPont have demonstrated the capability of patterning SCALPEL masks. Furthermore, Lucent Technologies and Beta Squared, a subsidiary of Photronics, recently agreed to jointly develop mask cleaning technology for the SCALPEL program.
The next step in the evolution of SCALPEL is to build a high throughput exposure tool. The biggest issues confronting researchers are the need for higher currents and addressing the corresponding exacerbation of space charge effects. Through collaboration with equipment manufacturers, the first commercial tools are targeted for availability as early as 2002, Harriott said.
IPL
|
|
Fig. 4. Resolution capabilities of IPL is demonstrated in the formation of 50 nm spaces imaged in Shipley's DUV resist. (Source: FhG ISiT, Berlin) |
Among the four NGL techniques, at least one study has cited IPL as having the lowest cost of ownership. Ion projection lithography resembles a cathode ray tube with multi-electrode electrostatic optics directing hydrogen or helium ions to the wafer. One tool design, developed at Ionen-Mikrofakrikations Systeme (Vienna, Austria) under a European MEDEA program and led by Siemens AG (Munich, Germany), replaces a full-field stepper with a stitcher concept to eliminate the need for new designs with each progressive increase in field size and resolution. Taking into consideration the focal length and beam half-angle dependence of blur, researchers determined that higher resolution can be obtained when a die is printed in smaller subfields (12.5 x 12.5 mm2). With a research ion projection tool at FhG ISiT (Berlin), feasibility of 50 nm resolution was demonstrated by using an exposure dose of 0.3 µC/cm2 (Fig. 4).
The stitcher approach will be the basis for all Siemens IPL tools down to 50 nm design rules and possibly lower with this year marking the availability of the industry's first IPL process development tool. Design and development have been completed, and parts currently are being manufactured. With the basic design already in place, transition to a viable production tool will be focused on 'fine-tuning' and improving the optical design of the system. According to Dr. Rainer Kaesmaier, technology champion from Siemens, throughput is anticipated to be 40 wph for 100 nm technology and 20 wph for 50 nm. Here, an advantage of the ion projection stitcher over electron projection systems may be realized, because the allowable current impacting beam blur need not be dramatically decreased in order to maintain throughput, Kaesmaier said. Also available this year is a stencil mask consisting of a 3 µm silicon membrane. Development was started a year and a half ago, and thus far, 80 masks have been produced. While the general design has been determined, work continues to eliminate distortions and non-uniformities.
Cost-of-ownership
Stepper throughput has traditionally dominated COO. Today, advanced optical lithography systems beyond 0.25 µm and all NGL tools will depend heavily on mask costs, which in turn will depend on usage. The average use of DRAM masks, for example, may be as high as 10,000 wafers, whereas for microprocessors, the average is 1500 and for ASICs ~250. Comparing NGL mask technologies, studies performed by International SEMATECH indicate IPL, SCALPEL and X-ray as having comparable mask costs at 100 nm design rules, with SCALPEL the lowest. With this in mind, SCALPEL may have the lowest COO, where mask costs are dominant (MPUs and ASICs), while IPL with relatively high mask costs but good projected throughput may be more suitable in DRAM manufacturing.
Each of the four NGL technologies vying to succeed optical lithography have notable strengths. EUV can potentially maintain high throughput while meeting linewidth requirements; X-ray has pre-production tools currently available and is proven technology, and SCALPEL and IPL may have COO advantages. Which technology will emerge as heir? Will more than one advance? Only time will tell. 
Challenges of Next-Generation Lithography
02/09/2010Revolutionizing Process Integration
02/09/2010Lithography: 0.18 µm and Beyond
01/31/1998
