Scaling studies of hot electron injection and interface-state generation in deep-submicron silicon mosfets: a monte carlo analysis

Abstract

The simulation model employed in this study consists of an advanced ensemble Monte Carlo method, that incorporates two conduction energy bands from pseudopotential calculations, coupled with an interface-state genemtion model. It has been demonstrated that this coupled treatment can calculate interface-state generation with accuracy and a good overall agreement has been achieved between the simulated results and measured data in longchannel (l-pm) devices [ I]. Using the same methodology, we explored the implications of power-supply voltages driven by two widely used device-scaling approaches: constant field scaling [2] 'and a more generalized scaling [3]. Throughout this study, the devices are stressed for 120 seconds at Vds=2Vgs. For constant field scaling, we found that the simulated electron injection rates decreased from 6.4x102'/cm2sec at the location of peak electron injection for a 0.33-pin device with Vd,=3.3 V, to 2.4x10"/cm2sec for a 0.12-pin device with Vd,=1.2 V. This corresponds to a predicted peak interface-state density of 7x10"/cm2eV and 4.3x10"/cm2eV for the 0.33-pm and 0.12-pm device, respectively. This decrease seems to be caused mainly by the reduced lateral field, (The "constant" lateral field methodology reduces the peak electric field from 170 kV/cm to 150 kV/cm for these short-channel devices.) On the other hand, the peak average electron energy is significantly reduced from 2.2 eV for the 0.33-pm device to 1.3 eV for the 0.12-pin device. When the generalized scaling scheme was applied, the simulated electron injection and interface-state generation rates increased considerably, from an interface-state generation density of 7x10"/cm2eV for the same 0.33-pm device with Vh=3.3 V, to 9.2x10"/cm2eV for a 0.12-pm device with Vd,=2.25 V. The corresponding peak electron injection rate for the 0.12-pm device was 1.1x1OZ2/cm2sec. For the generalized scaling scheme, the peak lateral field increased from 170 kV/cm to 240 kV/cm as the devices scaled down. The effect of power-supplyvoltage reduction was clearly seen in the average electron energy. The average energy at the location of pe'ak electron injection was significantly reduced, from 2.2 eV for the 0.33-pm device to 1.5 eV for the 0.12-pm device. Thus, the electron energy distribution appears to have strongly non-linear characteristics. While the average energy scales down with the power-supply voltage, the electrons in the high-energy tail of the distribution ciui bc enhanced (i.e., a longer tail) by the large peak electric field. These results for the two scaling approaches demonstrate the importance of hot electron degradation in deep-submicron MOSFETS operating below the 3 V power supply level.