2
\$\begingroup\$

Most undergrad textbooks on microelectronics (those covering semiconductors, diodes, BJT's, FET's, etc.) that I have laid my hands on discuss semiconductor concepts such as charge carries, drift, recombination, among others, to explain why transistors behave the way they do, and then present mathematical models for how different types of transistors work (e.g. IV curves) in order to analyse circuits involving them.

My question is, can state-of-the-art transistors such as those in modern CPU's be modeled by similar formulas? Given that modern transistors based on 7 nm technology can be only a few atoms in size (~30), do the same concepts apply, or do you really need PhD-level knowledge to even begin to understand how they actually work?

I've seen from other sources how ASIC foundries will share pspice models of the transistors that they can handle to customers and that they tend to be extremely complicated with dozens if not hundreds of parameters, so I got curious about this.

\$\endgroup\$
2
  • \$\begingroup\$ In the end even those concepts are just higher-level 'models' of the quantum-physics involved in a transistor structure. With such small transistors you can't really consider edges to be abrupt anymore which is why these models become quite complicated. \$\endgroup\$ – Joren Vaes Aug 30 '18 at 5:48
  • \$\begingroup\$ The pspice models have to be more verbose than just a textbook conceptualization of the physical processes. I bet that a regular jellybean discrete FET (e.g. BS170) also has a complicated pspice model (with dozens of parameters). \$\endgroup\$ – anrieff Aug 30 '18 at 5:52
2
\$\begingroup\$

My question is, can state-of-the-art transistors such as those in modern CPU's be modeled by similar formulas?

Yes

You need to understand what a model is.

A model is a description of a subset of the behaviours of something, that is sufficiently accurate for some stated purpose.

That is why you have different levels of model for the same thing, you have different purposes for using the model.

One model of a bipolar transistor is a IBE-dependent CE current source. This is enough for many purposes, crudely biassing a transistor for instance, or understanding a logic gate.

However, you need a more accurate model for building an amplifier, or a high speed logic gate. This is why you have Ebers-Moll, Gummel-Poon, behavioural SPICE models, measured S_parameters, and a few more besides.

Each is as simple as possible, while being good enough for the job in hand. None captures the entire range of behaviour shown by a real life transistor.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.