dc.description.abstract | The challenges for the production of SJ MOSFET lies in the exquisite design of charge-balancing. In this thesis, we systematically analyze the depth profile of E-field across p-pillars under various conditions including charge imbalance and charge balance. This allows us to gain insight on the E-field distribution, facilitating the design of p-pillar dosage, p-pillar width and n-epi concentration to achieve “charge balance” precisely and rapidly. Considering the process variations and following the charge balance principle, we have proposed a simple, effective design rule for downscaling the SJ MOSFETs with cell pitch of 1420m. The design rule is briefed as follows: (1) WP = WN = CP/2, (2) Poly Gate Width = 0.7CP, (3) achieving charge balance at the condition of WP × NP = WN × NN.
In addition to the device structure design in the cell, the structural design for the termination regions are also important in order to sustain high break down voltage for the entire system. We utilize two methods to make the E-field distribution as uniform as possible (to avoid the maximum E-filed occurring in specific regions). (1) A grading separation between pillar to pillar across the termination region, (2) Optimizing the field plate length design. We have used SILVACO Inc. Athena and Atlas simulator to simulate the device process and the corresponding electrical characteristics, respectively.
The effectiveness of simulation results has been verified by the experimental data from real devices based on our design. Decreasing the cell pitch from 20m to 16 m, experiment data demonstrate that, although the breakdown voltage declines from 724 to 678V, we are able to reduce the Ron,sp from 3.4 to 2.45 -mm2 and QG from 9.6 to 6.2nC simultaneously, effectively improving the FOM(Rdson×QG) approximately by 55%. For the further advancement of optimal FOM, we modulate the gate electrode width to reduce the gate charge. Simulation results show that decreasing gate electrode width from 11m to 9m, the FOM can be further improves by more than 20%.
| en_US |