The fast modeling method of MOS transistor electricity statistical model
Technical field
The present invention relates to a kind of fast modeling method of MOS transistor electricity statistical model.
Background technology
Integrated circuit (IC) products generally will be passed through roads up to a hundred process procedure in the technology manufacture process.Because the influence of the uncertain factor of each road technology on statistical significance, even product for same design, its circuit performance also can be owing to different manufacturing shops, the different batches of technology, different wafers and different chip positions and corresponding variation takes place.
Therefore, the device model of setting up when integrated circuit (IC) design should take into full account the influence of these uncertain statistical considerations, sets up the corresponding devices electricity statistical model.When utilizing the model of this foundation to do Monte-Carlo Simulation, analog simulation obtains the circuit performance statistical distribution should keep basically identical with the circuit performance statistical distribution that actual process is made.
At present, the method for setting up the MOS transistor electricity statistical model generally is divided into 2 kinds: propagated forward (forward propagation) and back-propagating (backward propagation).The major advantage of " propagated forward " method is that process is straightforward, but difficult definite all model parameter standard deviations.The major advantage of " back-propagating " method is that the model parameter standard deviation of determining has suitable reliability, but difficult point is the analysis of the sensitivity of process deviation and model bias.
Because " back-propagating " adopts the method from the reverse prediction model deviation of the process deviation that can survey, model bias has higher confidence level, thereby industry member is generally taked the modeling method of back-propagating.Then the difficulty to the statistical model modeling method of propagating then is the choosing method of statistical nature parameter and the analysis of associated sensitivity with key.Only choose suitable statistical nature parameter, calculate accurately associated sensitivity and set up statistical model in view of the above, the analog result that obtains through emulation just can match with the statistics of reality.
For setting up the MOS transistor electricity statistical model, at present, many companies have all developed the way that different statistical nature parameters chooses and the analytical technology of sensitivity.But these methods, the particularly analytical technology of sensitivity wherein often relate to complex mathematical computing and loaded down with trivial details numerical optimization.
The back-propagating method is after obtaining a large amount of device statisticss, the technological standards deviation can be found the solution from the analysis of data, the exploitation key of statistical model is the sensitivity of analysis and definite process deviation and model bias, therefore, adopt the matrix operation shown in the formula (1) to obtain the model parameter standard deviation usually.
Formula (1)
In the above formula,
Be respectively the technological standards deviation, the sensitivity of model parameter standard deviation and process deviation and model bias.
In the conventional at present MOS transistor electricity statistical model exploitation, the statistical nature parameter of generally choosing comprises Tox, Nch, Vfb, Dvt, K2, K3, K3B, Lint, Wint, U0, Xl, Xw, Vth0, Xj, Rsh or the like.Parameter is many, and when analyzing and determine sensitivity, need carry out complicated finding the solution according to the formula of MOS transistor physics analytic model.
The fast modeling method that the purpose of this invention is to provide a kind of MOS transistor electricity statistical model.
Summary of the invention
Technical matters to be solved by this invention provides a kind of fast modeling method of MOS transistor electricity statistical model.The characteristics of this method are, only choose the model parameter in the MOS BSIM4 model of these 6 industry member standards of threshold voltage vt h0 of the short-channel effect coefficient k 1 of the variation Xw of variation Xl, the channel width due to the technology of the channel length due to oxidated layer thickness Tox, the technology, piece resistance R sh that source electrode contacts with drain electrode, threshold voltage and device, and in sensitivity analysis, utilization numerical difference between point-score, by choosing suitable difference stepping, final definite rational Sensitirity va1ue, thereby the energy rapid modeling improves the efficient and the accuracy of integrated circuit (IC) design work.
The fast modeling method of MOS transistor electricity statistical model of the present invention, be characterized in that this method is included in the step of collecting a large amount of technological parameters in the production line and obtaining the technological standards deviation relevant with these technological parameters, choose 6 modeling statistics parameters and carry out model parameter and the step of the sensitivity analysis of technological standards deviation, the step of the standard deviation of the anti-launch mode shape parameter of on known sensitivity basis, collecting in by production line of technological standards deviation, and the standard deviation of model parameter write model file and carry out emulation and step that the standard deviation value of fine setting model parameter conforms to substantially up to emulation technological standards deviation and the data of surveying the technological standards deviation.
Use the present invention, can choose the characteristic parameter in the electricity statistical model quickly and analyze associated sensitivity, improve modeling efficiency.
The fast modeling method of MOS transistor electricity statistical model of the present invention, 6 model parameters choosing are oxidated layer thickness Tox, variation Xl, the variation Xw of the channel width due to the technology, the piece resistance R sh that source electrode contacts with drain electrode, the short-channel effect coefficient k 1 of threshold voltage and the threshold voltage vt h0 of device of the channel length due to the technology among the MOS BSIM4 of industry member standard.
This method, not to adopt the statistical nature parameter of generally choosing to comprise the complicated approach of Tox, Nch, Vfb, Dvt, K2, K3, K3B, Lint, Wint, U0, Xl, Xw, Vth0, Xj and Rsh, but adopt bigger 6 model parameters of device simulation Effect on Performance degree, these statistical nature parameters had both had physical property and measurability, had higher sensitivity again.Therefore, above model parameter is comparatively ideal statistical nature parameter, can obtain modeling effect fast and accurately.
The fast modeling method of MOS transistor electricity statistical model of the present invention can be in the sensitivity analysis step, and the method for utilization diff by the difference stepping of choosing, is determined reasonable sensitivity number.
In the analysis of sensitivity and when definite, give full play to the inner complete physics of HSPICE or SPECTRE emulator and resolve model bank and high-speed computing, utilization diff method, by choosing suitable difference stepping, can obtain more rational sensitivity number quickly, thereby obtain modeling effect fast and accurately.
The fast modeling method of MOS transistor electricity statistical model of the present invention, can make the difference stepping of choosing in the numerical difference between point-score is Δ Tox=0.1nm, Δ Xl=1nm, Δ Xw=1nm, Δ Rsh=0.01 Ω, Δ k1=0.01 and Δ Vth0=0.1mV, and the final reasonable Sensitirity va1ue of determining is followed successively by-1.62 ,-0.38,0.02 ,-0.1,7.045 and 0.5 to saturation current, and threshold voltage then is followed successively by 0.2,0.548 ,-0.0055,1.012 ,-0.2 and 0.4.
By choosing of above-mentioned difference stepped parameter, can obtain more rational sensitivity number quickly, thereby obtain modeling effect fast and accurately.
Description of drawings
Fig. 1 is a process chart of the present invention.
In the accompanying drawing: S1-technological standards deviation; The numerical difference between point-score sensitivity analysis of S2-6 model parameter and technological standards deviation; S3-model parameter standard deviation; S4-emulation technological standards deviation=actual measurement technological standards deviation.
Embodiment
The present invention is further detailed explanation below in conjunction with accompanying drawing.
Embodiment 1
As shown in Figure 1, the fast modeling method of MOS transistor electricity statistical model of the present invention, be included in the step S1 that collects a large amount of technological parameters in the production line and obtain the technological standards deviation relevant with these technological parameters, choose 6 model parameters and carry out model parameter and the step S2 of the numerical difference between point-score sensitivity analysis of technological standards deviation, the step S3 of the reverse launch mode shape parameter of the technological standards deviation standard deviation of on known sensitivity basis, collecting in by production line, and the step that the model parameter standard deviation is write model file and carries out emulation, and in this simulation process, also the model parameter standard deviation value is finely tuned the step S4 that conforms to substantially up to emulation technological standards deviation and both results of actual measurement technological standards deviation.
Embodiment 2
The fast modeling method of present embodiment 2 relates to the fast modeling method in the embodiment 1, and wherein 6 model parameters are the oxidated layer thickness Tox among the industry member standard MOS BSIM4, variation Xl, the variation Xw of the channel width due to the technology, the piece resistance R sh that source electrode contacts with drain electrode, the short-channel effect coefficient k 1 of threshold voltage and the threshold voltage vt h0 of device of the channel length due to the technology.
Embodiment 3
The fast modeling method of present embodiment 3 relates to the fast modeling method in the embodiment 2, wherein in the method for sensitivity analysis step utilization diff, by the difference stepping of choosing, determines reasonable sensitivity number.
Embodiment 4
The fast modeling method of present embodiment 4 relates to the fast modeling method in the embodiment 3, the difference stepping of wherein choosing is Δ Tox=0.1nm, Δ Xl=1nm, Δ Xw=1nm, Δ Rsh=0.01 Ω, Δ k1=0.01 and Δ Vth0=0.1mV, final definite reasonable Sensitirity va1ue is followed successively by-1.62 ,-0.38,0.02 ,-0.1,7.045 and 0.5 to saturation current, and threshold voltage then is followed successively by 0.2,0.548 ,-0.0055,1.012 ,-0.2 and 0.4.