MRS covers semiconductor topics
07/01/1997
MRS covers semiconductor topics
The Materials Research Society Spring Meeting, held in San Francisco, CA, covered a wide range of materials and processing topics. This report highlights those papers most relevant to semiconductor fabrication.
Amorphous silicon devices. Amorphous silicon is the most common substrate for active matrix liquid crystal displays (AMLCDs), yet its limited mobility constrains display resolution and transistor density. Masakiyo Matsumura and Akihiko Saitoh, of Tokyo Institute of Technology, offered an alternative design with vertical transistor structures. Gate metal (1?m Mo) was conformally deposited over a step and then etched by anisotropic RIE to leave metal on the sidewalls only.
Excimer laser selective crystallization of 50-nm phosphorus doped ?-Si forms active structures: Conformal ?-Si is deposited, then laser pulses (normal to the surface) crystallize the step top to form the source and the step bottom to form the drain. The ?-Si on the sidewall is not crystallized by the laser, so an isotropic etch easily removes it because of selectivity between ?- and poly-Si. Gate lengths (L) of 0.5 and 1.5 ?m were produced with W/L=100/1.5 and attractive short-channel characteristics.
At the other resolution extreme, Helena Gleskova and co-workers at Princeton and the University of Alabama demonstrated a remarkable use for an extra laser printer sitting around an office: LCD fabrication. Using toner etch masks (originally developed for printed circuit boards), they created functional low-resolution structures without the use of photoresist. Patterns were generated using computer software, and the first layer directly patterned onto thin glass sheets run through the laser printer (the back door has to be open so that the glass can run straight through).
The second and third masks were created by printing the toner onto transfer-paper, placing it on the substrate and aligning with an optical microscope, and then transferring the resist with heat and pressure (like a T-shirt print). A DI water bath lifts off the paper leaving the toner behind. They produced 50-?m lines using a 600 dpi laser-printer (? 10-?m particles) for toner printing; an improved toner with 1-?m particles could produce 5-?m resolution.
Defects and diffusion phenomena. A number of papers addressed transient enhanced diffusion (TED) effects. As
 |
Low temperature transient-enhanced diffusion effects for 5-keV arsenic implanted at a dose of 1015/cm2 and annealed at 650, 700, and 750?C.
transistor channel lengths are reduced, the vertical and lateral dopant distributions in the source/drain regions must also scale. Damage generated during ion implantation can cause TED of implanted regions, limiting the minimum junction depth. TED also modifies the doping profile of subsequent implants.
According to Intel Corp.`s Martin Giles, implant damage creation involves point and extended defects, dopant-defect clusters, and generation and recombination of cascades of defects. The final doping profile depends on interactions among multiple dopant types and multiple implantation and annealing cycles (see figure). These processes are not well understood, but a simplistic "+1" model has been useful in providing initial diffusion profiles. This model assumes that local interstitial-vacancy recombination consumes most of the Frenkel pairs generated in the damage cascade, leaving an interstitial profile dominated by silicon lattice atoms displaced by substitutional dopant atoms. The +1 model correctly predicts that two implants with the same dose but different total damage will have the same TED.
Alp Gencer, Iuval Clejan, and Scott T. Dunham at Boston University, examined the effects of extended defects in TED. They found that these defects store interstitials created by implantation, greatly prolonging the time period over which TED lasts.
David Venables of North Carolina State University and co-workers agreed. They examined the role of vacancies and intersititials in TED of arsenic using antimony and boron superlattices. Almost no diffusion occurred after arsenic implants into antimony superlattices (an interstitial diffuser), while implantation layers in boron superlattices (a vacancy diffuser) were almost completely washed out. The researchers concluded that vacancies play no direct role in TED.
Since junction depths need to be between half and one-quarter of the channel length, TED presents serious problems for 0.18-?m transistors. Several researchers demonstrated the necessary ultra-shallow junctions using very low (<1 keV) energy implants with rapid thermal anneals for activation. According to Aditya Agarwal, of Oak Ridge National Laboratory, most TED-induced profile movement occurs in the first 5 sec, so pre-amorphization of the substrate doesn`t necessarily help - the dopant can move further during the fast diffusion stage. At lower implant energies, though, defects are more stable.
Rapid thermal processing. TED is one of the reasons why thermal budget decreases at smaller design rules. RTP, which uses higher temperatures for shorter times, is one solution to the thermal budget crunch. Mike Pas from Texas Instruments reports that TI has moved source/drain anneals, silicide formation, nitrided gate oxide deposition, and BPSG reflow to RTP systems in 0.35-?m production. These steps represent only 15-20% of all thermal processes. More steps will be possible for RTP at 0.25-?m design rules because oxide layers will be thinner.
Further use of RTP is limited, Pas said, by cost of ownership concerns. RTP is not yet able to produce oxide layers more than 100 ? thick, and RTCVD suffers from temperature control, reliability, and throughput concerns. In response to a question about the risks of batch furnace processing for 300-mm wafers, Pas said that he saw "no issue" in committing large batches of wafers to a reliable tool.
Edmund Seebauer at the University of Illinois, Urbana, questioned the thermal budget concept itself, as it is usually understood. While thermal budget is an appealing metaphor, he said, it focuses on the initial and final wafer states, completely ignoring processes during the heating cycle. He proposed an alternative concept focusing on the activation energies of desired and undesired phenomena. This approach, he said, would allow temperature vs. time tradeoffs based on the relative rates of desired and undesired processes.
Organic electronic materials. Organic devices are slowly but surely gaining in manufacturability and complexity. Researchers at the University of Sheffield presented 60% quantum efficiency organic LEDs that were produced with polyfluorine source material from Dow Chemical (two benzene rings with alkyl chains, fully conjugated polymer, emission peaking in the blue).
Materials reliability. Of course, none of these advanced processes and materials matter if they don`t survive actual fabrication and use. As advanced technologies, such as copper, approach production-readiness, reliability becomes a more serious concern.
Greg Braeckelmann and coworkers at SUNY-Albany investigated the influence of undercoats/barrier layers on copper CVD adhesion. Common goals for this process are 2000 ?/min deposition rates in 0.18 ?m, 4:1 aspect ratio structures, with proper electrically conductive and chemically isolating barrier layers. Cu adhesion is critical for process compatibility with CMP and ultimate interconnect reliability. Ti, Ta, and W liners were studied.
For a TiN layer, air exposure before Cu CVD lowered adhesion (by approx. 4?); however, air exposed TiN is perfectly fine or even slightly better when used with PVD Cu. High power bias with PVD apparently implants the Cu through the surface contamination, thus allowing for excellent adhesion. Ar-plasma pre-treatment just before Cu CVD and after air exposure to TiN gave results similar to no air exposure. Hydrogen pretreatment didn`t help; it enhances nucleation density and surface roughness, but decreased adhesion strength due perhaps to "oxygen-contamination."
Haining S. Yang and co-workers at Rice University and Texas Instruments developed a new adhesion test, ELATE (ELectrostatic Adhesion Test), based on an electrostatically applied stress to parallel conductive films across a 2-?m vacuum gap (held constant by a ceramic spacer frame). When a film peels/blisters/fails, it typically delaminates from the substrate. Since the structure gap is small and a metal blister can only elastically deform to <1 ?m before failure, some portion of the blistered metal will contact the top electrode to short a circuit. Gradually ramping the applied voltage allows quantitative correlation to film stress.
Low-k dielectrics. The search for a viable, low-dielectric constant, alternative to SiO2 continues, with a variety of polymers being considered. IBM`s Almaden and T.J. Watson research centers presented work on organic-inorganic nanocomposites: poly(silsequioxanes) (PSQ) modified with new high-temperature polymer intermediates and molecular architectures. Pure PSQ is very brittle and >0.5-?m thick films are difficult to produce. Functional organic polymers "toughen" the film so it can be up to 2 ?m thick.
Nano-scale phase separation is important for gap-fill, and standard polyimides really only phase separate on the micron scale, not smaller. High boiling point solvents like di-methyl propylene urea are used to maintain miscibility until glass-phase formation "locks in" nano-phase separation. Approximately 20-23 wt% polyimide in PSQ, results in minimal wt% loss at 400?C and approx. k=2.7. A second complementary technique uses huge 3D "globular" organic molecules mixed with a more volatile second phase to get an open-cell porous "hyperbranched" structure with very low k.
Meanwhile, Kelly Taylor and co-workers at Texas Instruments used parylene copolymers, with nonpolar comonomers with separate phenyl groups, to lower the dielectric constant (k) below the 2.7 of standard parylene. However, just adding aliphatic carbons makes thermal stability weak, so CH2 groups were inserted between benzene rings. Additional additives include aromatic rings or fluorine to lower k, and Si to improve temperature stability.
The researchers used four different comonomer molecules: PFOMA, THPFUMA, APFB, and TV-TMCTS (tetra-vinyl "tomcats"). TV-TMCTS produces a k of 2.1 ?0.1 k with good adhesion and thermal stability: chuck temperature ?4 to ?10?C at 12-14 mtorr, with TV-TMCTS precursor at around 60?C. The four vinyl groups probably cross-linked to create nanovoids to achieve such a low, reproducible k. Though Si and SiO2 adhesion is excellent, Al adhesion is still a problem.
Workers at Hanyang University, National Institute of Technology and Quality, South Korea, reported on post-ECR deposition plasma-treatment of SiOF to improve moisture resistance. Surface roughness increased from 5 to 34 ? RMS (the entire range should be acceptable for devices). BOE etch rate decreased along an asymptotical curve from 80 to 20 ?/sec. RI and k increase to that of SiO2 after about 5 min, so the treatment is clearly removing F from the film (and somewhat densifying it). -E.K., K.D.