- 4.3.1 Piezoresistance Model
- 4.3.2 Physically Based Mobility Model for Strained Si
- 4.3.2.1 Electron Mobility in Unstrained and Undoped Si
- 4.3.2.2 Electron mobility in Strained and Undoped Si
- 4.3.2.3 Electron Mobility in Unstrained and Doped Si
- 4.3.2.4 Electron Mobility in Strained and Doped Si

- 4.3.3 Doping Dependence
- 4.3.4 Mobility under Shear Stress

4.3 Bulk Mobility of Strained Si

As was discussed in Section 3.3.3, the presence of mechanical strain modifies the relative position of the different valleys in the conduction band. A first order estimate of these effects on the mobility can be obtained using the piezoresistance model.

4.3.1 Piezoresistance Model

Piezoresistivity is the phenomena referring to the coupling between electrical conductivity (resistivity) and mechanical stress. Fig. (4.2) shows how the resistance of a n-type Si sample varies with hydrostatic stress. The resistance decreases linearly with stress until 20GPa. This change in resistance is related to a change in the resistivity, and can be expressed as

(4.33) |

The variation of the mobility with applied stress can be obtained from the piezoresistive coefficients originally measured by Smith [Smith54]. In the presence of stress, the conductivity of the unstrained semiconductor,

(4.34) |

gets modified to

(4.35) |

Here and denote respectively, the carrier concentration and mobility in the valley pair in strained Si. Assuming a doped semiconductor, ( ), the change in conductivity is given as [Manku93a].

(4.36) |

Here are the components of the strain tensor. The quantity is a fourth rank tensor of piezoresistance coefficients which are a measure of the first order change in normalized resistivity per unit applied stress for different stress directions. It has 81 elements which upon the application of the point group symmetry operations reduces to only 3 piezoresistance coefficients, , and . Table 4.3.1 lists the values of these piezoresistance coefficients for n-type and p-type Si. Using the contracted notation shown in (3.16), the change in mobility can be expressed as

p-type Si | |||

n-type Si |

(4.37) |

4.3.2 Physically Based Mobility Model for Strained Si

As suggested in [Manku92], the anisotropic electron mobility in strained Si can be computed as the weighted average of the unstrained electron mobility tensor, , of the conduction band valley pair in Si with the corresponding electron population, in the pair,

The relative populations of each valley pair is given by

where is calculated for non-degenerate doping concentrations using Boltzmann statistics with as the effective density of states.

The denote the strain-induced energy shifts which can be computed from deformation potential theory discussed in Section 3.3.1.1. The refer to the unstrained electron mobilities in the valley pair.

Here and denote the mobilities along the major and minor axes in each ellipsoid, respectively. The denote the electron mobility tensors for Si for the [100], [010], and [001] valley pairs corresponding to directions , , and , respectively. Using (4.38) to (4.41), the in-plane (x-component) and perpendicular components (z-component) of the electron mobility in strained Si on a (001) SiGe substrate can be expressed as

In arriving at (4.42) and (4.43), the relation has been used which is a consequence of the biaxial tensile strain resulting from growing Si on SiGe (see Section 3.3.3.1). The unstrained mobility can be obtained by setting .

To improve the electron mobility model for strained Si, the effect of inter-valley scattering is included. Equation (4.38) is modified as follows

Here the denote the electron mobility tensors of strained Si for the [100], [010], and [001] valleys pairs. In (4.45) a mobility tensor is modeled as a product of a scalar mobility and the scaled inverse mass tensor.

The scaled inverse effective mass tensors for the , and directions are given as

with and denoting the transversal and longitudinal masses, respectively for the ellipsoidal -valleys in Si. The mass tensors are scaled to a dimensionless form by the conductivity mass

From this scaling it follows that , where denotes the identity matrix. The scalar mobility includes the dependences on the energy shifts and the doping concentration of the strained Si layer.

In (4.49) the following momentum relaxation times are assumed:

- for acoustic intra-valley scattering and inter-valley scattering between equivalent valleys (-type).
- for inter-valley scattering between non-equivalent valleys (-type scattering).
- for impurity scattering.

Here denotes the inter-valley relaxation time for -type phonon scattering in unstrained Si. Since in the unstrained case all three valley pairs are equally populated, we have . Using (4.49) the total unstrained mobility can then be written as

Note that the sum evaluates to the identity matrix .

Here it is assumed that the strain-induced valley splitting is large enough such that the lowest valley is fully populated and the inter-valley scattering to higher valleys is suppressed.

The ratio of the fully-strained mobility to the unstrained mobility defines the mobility enhancement factor

where signifies the lattice mobility including the effect of impurity scattering. Multiplying the RHS of (4.54) with gives

Rearranging (4.55), we can express the ratio as

From (4.51), the lattice mobility can be rewritten as

Substituting the value of from (4.57) into (4.56) gives

The last relation in (4.58) is obtained using (4.53).

In strained Si, the total rate for electrons to scatter from initial valley to final valleys and is given by

For low electric fields an equilibrium distribution function can be assumed and can be calculated [Conwell67]:

with the inter-valley scattering rate defined as

(4.62) | |

(4.63) | |

(4.64) |

and the Boltzmann distribution function .

(4.65) |

Here denotes the phonon energy and the strain induced splitting, and C is a constant. A more accurate formalism for calculating the scattering rate would be through the incorporation of the density of states into (4.61), giving [Seeger88]

The calculation of the scattering rate using (4.61) and (4.66) is shown in Appendix A. For our modeling purpose, the definition in (4.61) leading to simpler expressions was found to be sufficient.

Using these expressions, can be expressed as

The function is defined as

Here and denotes the incomplete Gamma function. In the unstrained case

(4.69) |

and therefore the unstrained inter-valley relaxation time can be obtained as

The factor in (4.59) is thus determined from (4.67) and (4.70).

Multiplying the RHS of (4.49) with gives

Using the relations in (4.53), (4.57) and (4.58), the electron mobility for the valley in strained Si can be written as

where denotes the scaled effective mass tensor for the valley pair in (4.47) and . Equation (4.72) is plugged into (4.45) to give the total mobility tensor for electrons in strained Si as a function of doping concentration and strain. The tensor in (4.72) is given in the principal coordinate system and has diagonal form.

4.3.3 Doping Dependence

where, is the mobility for the undoped material, is the mobility at the highest doping. All other parameters are used as fitting parameters. Although initially proposed for the majority electron mobility in Si, equation (4.74) offers enough flexibility to model also the minority electron mobility in Si. The difference between majority and minority electron mobilities [Masetti83] is caused by effects such as degeneracy and the different screening behavior of electrons and holes in the semiconductor. Equation (4.74) describes a mathematical function with two extreme values and can deliver a second maximum or minimum at very high doping concentrations depending on the sign of . Thus, it allows both majority and minority carrier mobilities to be properly modeled.

- The valleys located along the and directions
move up in energy with respect to the valleys located along the
direction.
- The shape of the valleys located along the direction is distorted
which results in a variation of the effective masses.
- The band minima of the valley pair along the direction move towards the zone boundary points, .

The model presented in Section 4.3.2 relies on a) a model for the momentum relaxation time, b) the relative populations of the different valley pairs as a result of the energy shifts, and c) an effective mass tensor (with constant and ), which basically provides the tensorial description to the mobility. However, as discussed in Section 3.3.4, in the presence of a uniaxial tensile stress along the two-fold degenerate -valleys which are lowered in energy experience a change in the effective masses.

For the calculation of the mobilities in the presence of shear strain, the effective mass tensors in (4.47) have to be modified. Using (3.35) the energy dispersion relation for the lower lying valleys for stress along the direction can be rewritten as

Utilizing the transformation (3.66), the dispersion relation in the principal coordinate system () modifies to

(4.76) |

which can be expressed as

Here and the matrix denotes the inverse mass tensor,

(4.79) | |

(4.80) | |

(4.81) |

The transversal masses along the and directions of the -valleys are defined by and , respectively. The variation of these masses can be expressed as a function of strain as described in (3.73) and (3.74). Substituting (4.78) into (4.72) gives the mobility tensor which now has a non-diagonal form in the principal coordinate system.

The scaled inverse mass tensor in (4.78) was derived assuming only to be non-zero. For non zero values of the and shear strain components, the scaled inverse mass tensor in (4.78) can be permutated to obtain the scaled inverse mass tensors for the and valleys as

S. Dhar: Analytical Mobility Modeling for Strained Silicon-Based Devices