Modeling Of FinFET Dynamic Multiple Thermal RC Network And Its Nanosecond-level Electrothermal Coupling Characteristics

As the semiconductor industry developing along Moore's Law,continuously increasing chip integration brings higher power density,resulting in higher internal temperature of the integrated circuit,and the self-heating effect of the device becomes more and more significant.In advanced semiconductor technology,FinFET becomes popular due to its excellent ability to suppress short-channel effects.However,the complex device structure leads to complex heat dissipation paths and difficulty in heat dissipation,which intensifies the self-heating effect of the device,resulting in the device electrical performance degradation and reliability issues.Accurate spatiotemporal modeling of the self-heating effect has important research significance as it's not only optimizing the device model,but also helpful in circuit simulation and reliability design.Focus on the dynamic analysis of the self-heating effect of FinFET devices under the deep nanometer process node,the influence of device parameters on the thermal network,the size dependence of thermal resistance and capacitance,and the thermal coupling model of multi-fin structures are studied and discussed.The size-dependence5-step thermal RC network model is proposed,which has high accuracy in estimating the device temperature at nanosecond level.Embedding the model into the BSIM-CMG model improves the accuracy of temperature prediction at high frequencies.The main research and results are as follows:Firstly,a 5-step thermal RC network model is proposed to accurately characterize the transient temperature response of FinFETs.The 5-step thermal RC network with high accuracy in calculating the peak temperature of the device is extracted based on the structure function method by extracting the thermal time constant spectrum,network converting and constructing the structure function.Based on finite element simulation of 14-nm FinFETs' transient self-heating process by COMSOL,the average relative error of the extracted 5-step thermal RC network is 0.31 %,and the root mean square error is 1.73 K.Combined with the spatiotemporal heat distribution in the device,the thermal coupling study of the multi-fin structure is carried out based on the thermal resistance matrix.Secondly,the dimension-dependent 5-step thermal RC network model is proposed to solve the self-heating calculation of the device under different geometric parameters.The electrical characteristics of FinFET are simulated and calibrated by TCAD to study the influence of device parameters on the self-heating effect.The 5-step thermal network is extracted based on the electrothermal coupling transient temperature response and the structure function method,and the influence of channel height,channel width,shallow trench isolation thickness,source-drain extension region length and thermal conductivity of spacer on each order thermal characteristic are studied.Compared with the finite element simulation,the one-step self-heating model in the BSIM-CMG has root mean square error of 15.66 K,while the proposed five-step thermal network dimension-dependent model has an average relative error of 0.16 %and the root mean square error is 0.73 K,which is 95.32 % lower than the one-step model,and has higher accuracy in predicting the self-heating effect of the device.Thirdly,we propose a precise electrothermal coupling co-design method which is compatible with BSIM-CMG to study the influence of self-heating effect on circuit performance.Extracting the device parameters based on the BSIM-CMG,electrothermal coupling simulate with BSIM-CMG model and the five-step thermal RC network model by adding thermal nodes.The effects of self-heating effects on inverters and ring oscillators are studied,and it is found that the DC gain of the inverter degrades by 12.02 %,the noise margin degrade by 6 m V and 2 m V,and the propagation delay and transition time degrade 2 ps and 1 ps respectively under the influence of selfheating effects.The propagation delay is positively related to the load capacitance.As the inverter stages increases,the temperature rise of the first cycle in ring oscillator increases due to the longer period,and the steady-state temperature rise decreases due to the reduction in dynamic power consumption.As the length of the source-drain extension region increases,the temperature rise of the seven-step ring oscillator decreases by about 7.22 %.The proposed multiple thermal network dimension dependent model and the selfheating analysis solving the accuracy degradation problem of the existing BSIM-CMG self-heating model at high frequencies,is of great significance for FinFET device model optimization and circuit electro-thermal reliability design and able to applied to the precise design of nanosecond-level circuits.