Loss function

AI Discussions
1

How to write the loss function for this paper can someone write it for (I-V)?? Also I have Id-Vgs data for which I have find three smooth differential di/dvg

Neural_Network-Based_and_Modeling_With_High_Accuracy_and_Potential_Model_Speed.pdf (2.0 MB)

1 Like
2

The problem here is that I am not a EE or Device Physics person, so it would take a lot more work for me to understand the point of what they are talking about there. The actual formulas they give look pretty straightforward, but the question is what is the actual data that you have to work with? Which variables do you actually get as inputs and in what form are they? Arrays or tensors of what shapes? And what is the form of the outputs of your model?

Note that some of the stuff they quote may look fancy, but is really just the basic definitions of logarithms and exponentials:

If we have:

y = log(\displaystyle \frac{I_{DS}}{V_{DS}})

z = e^y

That just says that:

z = \displaystyle \frac{I_{DS}}{V_{DS}}

So it shouldn’t be any big revelation that:

I_{DS} = V_{DS} * z

The other thing to note is that RMS I’m guessing means “root mean square”, as in:

RMS(v) = \displaystyle \sqrt{\frac{1}{n}\sum_{i = 1}^n v_i^2}

So then the question is back to what I asked at the beginning: what is the actual data that you have when it comes time to compute the loss? Where do you get things like \displaystyle \frac {\partial z_{true}}{\partial V_{DS}} and \displaystyle \frac {\partial z_{pred}}{\partial V_{DS}}? Or is this just more fancy handwaving like the stuff I showed above, since z = \displaystyle \frac{I_{DS}}{V_{DS}}?

2 Likes
3

MOSFET.pdf (263.8 KB)
I have attached a pdf a short description of MOSFET ,you will need to understand
also the problem is also written. if you have any further questions you can ask, but please help me as I trying to implement it wanted to apply its loss function so badly

If you agree we can have a quick meet on google meet where I can clear all your doubts it will take very less time

But please help me to write its loss function

Thanks

4

I took a quick look at the PDF and it’s just giving device physics information. I don’t think that sheds any additional light for the purpose of building a DL model. My question is what does the data actually look like. You already have the input data, right? How is it packaged? What are the sizes and shapes of all the arrays or tensors? That’s the level of information we need to figure out how to approach writing the code. We have the formulas. It doesn’t do me any additional good to know what V_{GS} actually represents at the device physics level. I want to know what your actual input V_{GS} data looks like. Is it in timesteps and multiple samples?

Also just at a high level we need to talk about the basic model here. We are all just fellow students and participants here. Nobody is paid to answer questions here, right? You don’t just get to assign your work to someone. You have to get them interested enough to help for no compensation other than the interest of it.

5

Mosfet_alt_new1.ipynb (10.2 KB)
nmos45idvsvgs.csv (86.6 KB)
hers is my code and data which I tried to have but the results I am getting are not good. Can you tell me what the problem it may has??

6

Also Sorry for late response , and I am not trying to assign you work I am also just wanted to learn, may be I just carried away with emotions of not solving the problem too much…