|
The Fine Art of Modeling
|
||||||
|
Got you didn't I? ;) Well, I do model...but fortunately (or unfortunately) I model semiconductor devices. OK....your eyelids are already drooping, I can see that. But then, it is not that boring. In fact, it can get pretty interesting!! Here is a low down on my research. I'll bet you want to know more once you read it. (In that case, I'll be glad to enlighten you more...my contact info is here.) If you'd like to visit my lab web page, here is the link.
FAQs on Semiconductor Device Modeling (1) What is semiconductor device modeling? Why do we model? Ans: If you have done Electrical Engineering, at one point you must've surely used/heard of PSPICE. You drag and drop passive elements (such as resistors, capacitors etc) and semiconductor devices (such as Diodes, MOSFETs, BJTs...) to make a circuit that maybe looks like this:
In the schematic above, the BC546 is the "MODEL" of a BJT. Hmm...so that gets you thinking...if the figure above is the 'Model' of the BC546, then it should behave like a BC546. So, that means somebody has studied how the BJT works and replicated the working through some means (shall we say, 'code'). That somebody will be a person like me, a modeler (in some dark, dank corner staring at his computer screen 18 hours a day :)) ). Therefore, a modeler created Q1 in the above figure with his 'code' also known as a 'Model'. Simple ain't it? So, the process of modeling can be summarized by the figure below:
Imagine a life without models...circuit designers have to fabricate/assemble each and every circuit even if they aren't sure if it would work or not! The time and cost involved for such an exercise would be stupendous and most of the circuits/systems would never see the light of the day, given the increasing complexity of circuits we currently build!! (2) How is the "code" written? Ans: Simple answer is--->there is no simple answer (no....tongue not in cheek). PSPICE models are written on a 'C' platform with many simulator specific information. It can get messy and definitely not the easiest of ways. So, we now have what is called Hardware Description Languages (HDLs) that simplify the way we write models (what are engineers for if not for simplifying things....hehehe). The models have mathematical relations (such as algebraic differential equations) and these relations can be directly represented using HDLs just like the way we do in Calculus I/II. This makes the whole process intuitive and models can be written in a jiffy (that's like a couple of months actually!). (3) So, what are the HDLs available? Ans: Some of the popular ones are (for Analog modeling): Verilog-A, VHDL-AMS, MAST etc. and each of them have an associated simulator. I model in the MAST HDL and simulate in the Saber simulator (from Synopsys Inc.). (4) What are these Silicon Carbide devices that I keep hearing about? Ans: Silicon has been the overwhelming favorite for making devices for the last 40 years are so. Device developers are coming to understand that Si is fast reaching it's limits with regards to power handling and thermal issues. Of course, we are talking here of 'Power' semiconductor devices. Si and the IC industry is a different story altogether. Silicon Carbide, a wide band gap material (3.2 eV) has been touted as a possible successor to Si in the power area. The reasons are pretty obvious. (a) Order of magnitude higher breakdown field than Si (b) Three times the saturation velocity of Si and (c) 3.5 times the thermal conductivity. The advantages of therefore pretty obvious: SiC can handle much higher blocking voltages and have a higher current density. Moreover, higher thermal conductivity obviates the need for cooling systems and therefore SiC based electronics can be deployed in high temperature environments. NASA has shown an SiC LED working at 600oC (picture below) and in our own lab at Arkansas, we have demonstrated a SiC based DC-DC buck converter working at 400oC.
The icing on the cake is the fact that SiC is also rad-hard! So, in all probability we will soon be seeing SiC based circuits used in missions sent to other planets. SiC devices also have a whole lot of applications in good ol' earth. For example, fighter aircrafts having SiC electronics for their control will be lighter as the electronic modules can be placed anywhere (near the engines too) and no cooling systems will be needed. Hybrid electric vehicles and the utility grid are other promising areas where SiC devices can be used to improve system performance. (5) Wow! They sound so cool. Can I pick up some SiC devices from Radio Shack? Ans: Nopes. Most of the SiC devices are still under development. SiCED (Infineon), SemiSouth and Cree are some of the players who have commercially launched SiC devices. Devices currently available are Schottky barrier diodes and JFETs. These are still pretty expensive. And that is because low defect SiC wafers are still too expensive for mass production unlike Silicon. This will remain so at least until growth technology drastically changes. (6) Are you modeling these new age devices then? Ans: Yup! you got that right. To fully harness this technology, power electronic designers need models that accurately predict the way these devices work. And that is what our lab is spearheading. We have developed models for SiC diodes, MOSFETs and JFETs. We are in the process of developing models for other power devices. (7) Can you elaborate a little more on the modeling methodology? Ans: OK. We discussed on writing the model code in a HDL. After that is done, we need to make sure that the model we developed actually behaves like the device. For this, we enter into a new realm called 'characterization'. In other words, we test the actual device and get 'characterized data'. This can be in the form of I-V curves (DC characteristics), switching waveforms (Transient characteristics), reverse recovery waveforms etc. Building test benches for testing these devices is an art by itself and I am not discussing it here. Now, the model that we created is 'validated' with the characterized data and seen if it matches. It should. If it does not, then we need to go back and tweak the model by varying its 'parameters' and try again. This is an iterative process that continues till the model matches with the data. After that, we have a 'Validated Model' ready to be used for simulation! The whole process can be summarized in the figure below.
(7) This sounds very interesting. I have a few questions. Can I ask you? Ans: Sure you may. My contact info is here. If you'd like to read some of my publications, drop me an email. |
||||||
Courtesy:
NASA Glenn