As silicon transistor dimensions scale down, transient enhanced diffusion of n-type dopants has become a major barrier toward achieving required junction depths for transistors. In this paper, we use density functional calculations to identify a pathway by which silicon interstitials can mediate As and P diffusion. We show that As-silicon interstitial and P-silicon interstitial pairs in the neutral and negative charge states diffuse via a mechanism in which the dopant is bond-centered at energy minima and threefold coordinated at the high energy saddle point during dopant migration. For both As-silicon interstitial and P-silicon interstitial pairs, we conclude that the neutral pairs will dominate under intrinsic conditions while the neutral and negatively charged pairs will both contribute under heavily doped extrinsic conditions.

As the fabrication for the CMOS technology trends to decreasing the size of transistors for the benefit of increased performance and device density, the implant and annealing process technologies are being forced to meet increasingly challenging demands for ultra-shallow junctions. As the current annealing preferences shift from rapid thermal annealing toward sub-second RTP technologies such as flash and laser annealing, the dynamics of micro-structural evolution on the order of milliseconds and microseconds become of increasing importance. To investigate the kinetics in this ultra-fast annealing regime, this study investigated the effect of varying the temperature and time of a scanning laser anneal in the millisecond and microsecond timeframe. Amorphous layers were created in silicon substrates using a 30 keV silicon implant at a dose of 1x1015 ions/cm3. These wafers were then processed using a scanning continuous wave near infra-red laser at varying temperatures and scan rates in order to vary the dwell time of each condition. A combination of sheet resistance, optical interferometry, and secondary ion mass spectrometry characterization techniques were used in order to investigate the activation and diffusion of these laser annealed samples. The regrowth temperature across all samples showed a positive correlation to the sheet resistance; trending toward lower values as temperatures increased. Dwell time, however, showed an inverse effect at lower temperatures. The microsecond anneal at 1200 °C showed a lower Rs than the millisecond anneal, while at higher temperatures the trend reversed. This trending shows an obvious competition between activation and diffusion mechanisms and their dependence on regrowth temperature.

In order to form the deep PN junction demanded for the next generation IGBTs, the double-pulsed laser annealing technique as the low-thermal budget heat treatment has been introduced to activate a B-implant layer and a P-implant layer within the wafer surface to the depth 2μm. The double-pulsed laser annealing is characterized by the deep penetration depth due to a green wavelength of DPSS lasers and precisely and widely controlling of the annealing temperature and time. In the IGBT's structure the deep PN junction at a collector (the rear face) should be formed without damaging thermally circuit elements made of low melting point materials at a gate and an emitter (the front face).

Ion-implant samples using eight-inch (100) Si wafers were prepared as follows: Boron (B) implant was performed at a dose of 1E+15/cm2 at an energy of 40keV and/or phosphorus (P) implant at 1E+13/cm2 at an energy of 400keV. The double-pulsed laser irradiation was carried out at the constant first and second pulse energy density E1=E2=1.8J/cm2 at the delay time td= 0-500ns and the overlap ratio OR=67-90%. The melt depth was up to about 0.3μm. The electrical activation ratio of the B-implant layer within the depth of about 0.6μm was improved from 91% to about 100% with the delay time increase of 0ns to 500ns. The activation ratio of the P-implant layer within the depth of about 2μm was drastically improved from 48% to 82% with the same delay time increase and the carriers in the P-implant layer were distributed deeply from the depth 1.1μm to 1.8μm. Furthermore, with the overlap ratio increase of 67% to 90% the carriers in the P-implant layer were distributed deeply from the depth 1.8μm to 1.9μm and the high activation ratio of 82% was maintained. The high ratio of electrical activation is supported by the defect-free epitaxial regrowth where the majority of the B dopants was diffused in the liquid phase and that of the P dopants in the solid phase.

In this overview we examine how advanced continuum and atomistic modeling can help to understand and resolve process and device design issues for the 65 nm technology generation and beyond. The following implantation-related issues are reviewed: wafer temperature for different types of implant equipment and its impact on defect formation and amorphization, ion scattering off the photoresist mask and its impact on threshold voltage variation, dual rotation halo implant instead of the conventional quad rotation halo implant, and engineering of the source/drain junction overlap for diffusionless annealing by using tilted implants. The following annealing-related issues are also considered: limitations of spike anneal; benefits of cocktail junctions, heat transfer mechanisms for spike and millisecond annealing, and implant damage evolution for different thermal budgets. Taken together, implant, annealing, and layout conditions are shown to explain observed threshold voltage and transistor performance variations. In addition, the effects of transistor geometry on dopant diffusion, activation, and defect formation are shown for several generations of bulk and FDSOI MOSFETs.

We present B junction extensions that are extremely abrupt and shallow manufactured by amorphization, C co-implantation and conventional rapid thermal annealing (RTA). Resulting junctions have abruptnesses of 2 nm/dec better than as-implanted profiles. The most shallow B junction that has been manufactured is 15 nm deep and Rs = 626 Ω/sq. Successful implementation of these junctions is straightforward for P-MOS 30 nm gate length devices.

As millisecond annealing is increasingly utilized, the as-implanted profile dominates the final dopant distribution. We characterized boron diffusion in amorphous silicon prior to post-implantation annealing. SIMS confirmed that both fluorine and germanium enhance boron motion in amorphous materials. The magnitude of boron diffusion in germanium amorphized silicon scales with increasing fluorine dose. Boron atoms are mobile at concentrations approaching 1x1019 atoms/cm^3. It appears that defects inherent to the structure of amorphous silicon can trap and immobilize boron atoms at room temperature, but that chemical reactions involving Si-F and Si-Ge eliminate potential trapping sites. Sequential Ge+, F+, and B+ implants result in 80% more boron motion than do sequential Si+, F+, and B+ implants. The mobile boron dose and trapping site concentration change as functions of the fluorine dose through power law relationships. As the fluorine dose increases, the trapping site population decreases and the mobile boron dose increases. This reduction in trap density can result in as-implanted “junction depths” that are as much as 75% deeper (taken at 1x1018 atoms/cm-3) for samples implanted with 500 eV, 1x1015 atoms/cm2 boron.

During the MOS transistors fabrication process, the source-drain extension areas are directly in contact with the oxide liner of the spacers stack. In previous works [1, 2, 3] it has been established that boron can diffuse from the source-drain extensions into the spacer oxide liner during the subsequent annealing steps, and that the amount of boron loss depends on the hydrogen content in the oxide, because it enhances B diffusivity in SiO2.

In order to characterize and quantify the above phenomena, we performed test experiments on full sheet samples, which mimic either BF2 source-drain extensions over arsenic pockets implants, or BF2 pockets under arsenic or phosphorus source-drain extensions implants. Following the corresponding implants, the wafers were covered with different spacer stacks (oxide + nitride) deposited either by LPCVD, or PECVD. After appropriate activation annealing steps, SIMS measurements were used to characterize the profiles of the various dopants, and the corresponding dose loss was evaluated for each species.

Our experimental results clearly evidence that LPCVD or PECVD spacer stacks have no influence on the arsenic profiles. On the other hand, phosphorus and boron profiles are affected. For boron profiles, each spacer type has a different influence. It is also shown that boron out-diffuses not only from the B doped source-drain extension in direct contact with the oxide layer, but also from the "buried" B pockets lying under n-doped source drain extension areas. All these results are discussed in term of the possible relevant mechanism.

Ultra-shallow B and BF2 implants in silicon pre-amorphised with Ge have been activated using a scanning non-melt laser. The implants were activated either by using 1 or 10 laser scans. Isochronal 60s post-laser annealing between 700-1000°C were then undertaken to study the deactivation and reactivation of the B. Both B and BF2 samples were implanted with a dose of 1x1015 B cm-2 at an effective energy of 500eV. The presence of F from the BF2 implants, which is superimposed over the boron profile increases the sheet resistance of the initial fabricated junction (from 600-700 Ω/sq from B implants only to 750-1100 Ω/sq for BF2 implants). Fluorine also changes the deactivation and reactivation behaviour of the boron during the post-anneals by increasing the amount of deactivation of the boron.