Page 390 - Sami Franssila Introduction to Microfabrication
P. 390

Moore’s Law 369



           additionally, ions or electrons lead to mask charging and  to grain boundary reflections, and at 100 nm linewidths,
           damage. Electron scattering masks, instead of absorbing  copper resistivity has been reported to increase to
           masks, have been developed for EPL. This eliminates  4 µohm-cm.
           many of the thickness, stress and heating problems. Still,  Film thickness downscaling at the back end is driven
           at an estimated 15 million-dollar price tag, EPL systems  by the need to keep aspect ratios reasonable, even
           will only write 15 wafers per hour.         though RC-time delays inevitably increase as resistance
                                                       increases in thinner wires, and capacitance increases
                                                       when dielectrics are scaled down. Ultimate limits are
           38.5 FUNDAMENTAL AND PRACTICAL LIMITS
                                                       fairly close in back-end scaling: copper is as close to
           38.5.1 Linewidth and film thickness          minimum resistivity as any metal can practically be,
                                                       and with dielectrics, ε = 1 (vacuum) is not so far away,
           Nominal or design width is just an idealization of a
           microstructure. The physical structure in silicon or in  with ε = 2 materials being introduced. Superconducting
           thin-film material adds its own features. These effects  wiring was touted in the early 1990s as a solution to the
           are more pronounced the narrower the linewidth or  resistance problem, but enthusiasm waned rapidly when
           the thinner the film. The smaller the details we study,  the difficulties of a high-T c superconductor deposition
           the more are the effects that come into play. Line  and structural control became apparent.
           edge roughness can become significant when compared  Scaling to atomic dimensions leads to inevitable
           with linewidth. In the extreme, it is partly a materials  limitations. Gate oxide thickness is approaching such
           limitation: chrome, photoresist and thin film on wafer are  limits: because atoms are discrete, gate oxide thickness
           granular to some extent, and for instance, polycrystalline  is ‘quantized’ (Figure 38.4): we cannot have any gate
           materials may be etched at slightly different etch rates  oxide thickness, only integral multiples of atomic
           for different crystal orientations, and this preferential  dimensions. Putting it another way, each transistor will
                                                       have its own microscopic oxide thickness pattern, and
           etching contributes to line-edge roughness.
                                                       consequently idiosyncratic microroughness that affects
             In TiSi 2 formation on polysilicon, three-grain bound-
                                                       channel mobility and tunnelling currents.
           aries are crucial for nucleation of the C54 phase, but if
           the linewidth is narrow and the grain boundaries sparse,
           nucleation is retarded. This can be battled by increasing  38.5.2 Device considerations
           the annealing temperature, but this is at odds with dif-
           fusion goals, and it will also change the relative rates  When MOS transistors are made extremely small, the
           of silicidation and surface nitridation. Polysilicon grain-  ability of the gate to control the current in the channel is
           size tailoring by ion implantation before titanium deposi-  diminished. This can be overcome if two (or more) gates
           tion can be performed or alternatively the titanium depo-  are to be used instead of one, as shown in Figure 38.5.
           sition process can be modified by, for example, heating,  Fabrication of these devices is not obvious, and the two-
           or a thin (∼nanometre) intermediate layer of molybde-  gate version can exist in various configurations, with the
           num can be deposited between titanium and polysilicon  gates parallel to the silicon surface or vertical.
           to modify nucleation kinetics. Yet another method is  So far, very little attention has been paid to the
           ion beam mixing: the interface between the poly and  MOSFET channel, but of course, the channel can be
           titanium is modified by ion implantation after metal  improved and tailored just like gate oxide or junctions.
           deposition. The maximum projected range should coin-  Strained silicon is an actively studied channel material.
           cide with the film interface for maximum modification.  As discussed in connection with thin-film stresses,
             When linewidth scaling is continued, the relative  Si 1−x Ge x alloys have lattice constants larger than silicon
           importance of physical effects changes. Current con-  and they are under compressive stress, and consequently
           duction in a 1 × 1 µm cross-sectional conductor line is  the silicon on Si 1−x Ge x will be under tensile stress. This
           fully characterized by classical ohmic description. Nar-  tensile stress introduces energy split in the conduction
           rower lines and thinner films reach a limit at which  band of silicon, which leads to mobility enhancement,
           the surface scattering contribution to resistance becomes  for electrons by a factor of 2 and for holes by a factor
           important, and in the 10 nm-size range, quantum effects  or 4 (depending on germanium content, doping level
           come into play and single electron conduction can be  and field strength). Higher operating frequency could be
           seen. The characteristic scale for non-classical effects is  obtained from MOSFETs without lithographic scaling
           the mean free path, which is 40 nm for copper and 15 nm  (Figure 38.6).
           for aluminium. However, some deviation from classical  The smallest MOSFETs fabricated to date, have 6 nm
           behaviour has been seen even at 500 nm, probably due  gate lengths, and simple ring-oscillator circuits with
   385   386   387   388   389   390   391   392   393   394   395