If I want to design a chip myself, where to start with? I am a electrical engineer.

 

1. Choose a Fab.
2. Contact them about the process.

 

If I want to design a chip myself, where to start with? I am a electrical engineer.

Once you determine how many devices you need, speed, power, the volume you expect to sell, etc. contact one of the foundries to find out what process technologies would be appropriate, what the costs will be, and what tools are compatible with their PDK.

If you don't how to do physical layout, you'll have to contract someone who does so they can give you an area estimate based on your design.

Depending on your design, you'll need to do performance and reliability verification. The person who can do physical layout should be able to do layout and netlist verification.

Then you're going to need to have people who know how to make the frame. A foundry likely has tools to do this and may do it for you. I only worked for companies that had their own in-house manufacturing capabilities and provided documentation or assistance in adding alignment marks.

 

contact one of the foundries

I couldn’t think of that term when I said fab. I love it. I see huge vats of glowing molten metal being poured into a mold.

 

Start with an FPGA platform like xillinx.

 

Start with an FPGA platform like xillinx.

This. Exactly this.

 

Analog or digital IC?

 

1. start by putting down some numbers such as specs, performance, standards/approvals...
2. see if suitable candidate already exists. with so many products on the market there is a good chance that something suitable already exists.
3. if the functionality can be implemented in digital logic, then FPGA is a way to proceed. but... that will not work if you re building something analog like current or voltage reference etc. which is why you should not skip step 1.

 

Hello,

see that link:

https://tinytapeout.com/

BTW: there are many courses related to designing custom ASIC chips on the Udemy platform:

https://www.udemy.com/course/asic-b...filter_type=subs_only&couponCode=KEEPLEARNING

Best Regards

 

Really, the very first step in an integrated circuit design project is to determine the function of what is to be designed.
Will it be a custom digital watch or a new "quad channel audio processor with zero distortion", or maybe a language translator for an in-ear conversational device?? Or possibly a better op-amp IC.
After the function is decided then comes determining how the logic will work to provide that function. First the string of basic blocks, and then the longer string of functions, and the pages of circuits to produce those functions.
After that all you need to do is decide what technology is best suited, and then contact the appropriate Fab-House as described in posts #2 and #3.

 

If I want to design a chip myself, where to start with? I am a electrical engineer.

Hi,

As others have said, there are a lot of ways of doing custom work these days without designing an entirely new die.
Unless you need something very unique, for a larger production run, or it's for academic purposes, these other techniques could be much cheaper and faster.

I'd be curious to know what you want to design a new chip for.

 

If I want to design a chip myself, where to start with? I am a electrical engineer.

What do you mean by "design" a chip yourself?

Design the circuits that will be implemented in the IC?

Layout the IC artwork used to fab the chip?

What kind of chip? Analog? Digital? Mixed-signal? Sensor? Readout?

 

I couldn’t think of that term when I said fab. I love it. I see huge vats of glowing molten metal being poured into a mold.

The 'foundry' is the place that essentially takes the raw materials and produces the bare wafers.

The 'fab' is the place that takes the bare wafers and processes them to produce ICs.

The terms get mixed up a lot, including by those working in the industry, because few people interact with both, so when they use either term they are referring to the one they interact with. Context is usually sufficient to overcome any ambiguity.

 

Companies that make steel have foundries
Companies that use steel to make cars are the fabricators or assemblers of cars.

 

The 'foundry' is the place that essentially takes the raw materials and produces the bare wafers.

TSMC is the World's largest foundry. AMD spun off their fab and called it GlobalFoundries. Intel calls their foundry Intel Foundry Services. None of them make bare wafers.





Before Intel split itself into two "companies", the internal factories were called fabs.

 

TSMC is the World's largest foundry. AMD spun off their fab and called it GlobalFoundries. Intel calls their foundry Intel Foundry Services. None of them make bare wafers.

Before Intel split itself into two "companies", the internal factories were called fabs.

As I said: The terms get mixed up a lot, including by those working in the industry

I used to talk to people at TSMC, IBM, MOSIS, Tower and others routinely (we designed one of TSMC's early standard digital libraries for doing mixed-signal designs on their standard logic process and helped Tower develop their stitching technology because of the physical size of some of our designs) and one several occasions I brought this up in idle conversation. Sometimes the person didn't know and didn't care about the distinction, but when I was talking to someone that did, the response was always the same -- the terms are hopelessly intermingled in the industry to the point that companies will even knowingly misuse "foundry" and "foundry services" simply because that is the term that so many of their customers a used to and there is no point in trying to correct them. The fab manager at IBM that I spoke with pointed something else out -- the people in these organizations that decide was a business unit gets named seldom come from a technical background and they neither know nor care about the distinction, they simply want a name that they think will work best from a sales and financial reports standpoint.

 

The reality is that there is a vast amount of effort involved with the creation of an integrated circuit, even if one has the resources available. The steps I described would be a fair portion of what would be needed to produce an IC design to hand off to an existing IC manufacturer willing and able to add a new product.

Now, once again, the question of what sort of function would some new design provide? And would there be thousands of people willing to pay $10 per device for it?

 

Like emiconductors and how transistors work (MOSFETs, since you're involved with them).

I think semiconductor theory is not needed.
For digital ICs you really don't care how a AND gate is made. You just need that function and call out a AND gate.
For analog most ICs are designed by using building blocks that you do not get to make changes to. TI has a big list of analog blocks like a PGA103 amplifier building block. It just works and you wire it up in the compiler. Pin52 connects to the output.
The compilers I used does not care if you know what a transistor is or not.

I know there is a level where you can work at the transistor level but very few people do that.

Get familiar with digital logic design and VLSI

For the digital, it is done in a high-level logic like VLSI. You are not really calling out gate level functions like OR gates, but you are calling out functions. The computer decides how to make that function work.

If you finish the chips design, then found a Semiconductor manufacturer to help you make it.

I started with the manufacturer and used their tools.

 

Does anybody on this thread have an idea of what it would cost to develop a low-complexity gate array these days, and the minimum order to get fab & production house interested?

 

The last price that I heard, which was about 30 years ago, was about $30,000, IF the design was simply stitching together standard logic blocks to provide custom functionality in a 16 pin dip package, done by an organization that specialized in doing just that. And that price was only for repackaging some 4000 series IC functions into a single package.
Today that price would cover possibly one week of the creation process.

Go back to post#10 and review the steps in my very simplified description of the process.

The one part I overlooked was the research to see if there is already an IC available that duplicates the intended functionality.