Quote from: FinalHash on September 03, 2014, 09:29:57 PM

Great questions.

Let me start of first by laying out two scenarios for you.

Scenario 1: Making your own chips

Firstly, the name manufacturer may be a bit of a misnomer.

The process in ANY chip production life cycle is as follows.

Step1: Find a really smart person to make you what is known as an RTL http://en.wikipedia.org/wiki/Register-transfer_level
This is essentially the ground work for telling the chip what you want it to do. Think of it as sort of the firmware of the chip.
RTL is generally either written in VHDL or verilog.

Step2: Once you have your RTL completed, you need to compile it to an FPGA. Now I am sure you all remember the FPGA days of bitcoin machines. FPGA's in a general sense are just prototype boards that you can plug in your RTL into in order to make sure it works. So in the case of a crypto RTL, generally folks compile their RTL on something like a XILINX FPGA so they can validate their design. This is why you get many asic companies showing off their FPGA boards at the beginning stages. This illustrates, generally, proof of concept. So if their RTL works, they can partially compile a small number of cores to execute their RTL and in the case of crypto, these RTLs mine coins. So if your fpga mines or even accepts shares, you are golden from an RTL standpoint.

Step3: Initiate backend design. Now this backend design can constitute many things. Whether that is making the PCB design, or makeing the mask design. Essentially it is at this point that virtually all ASIC manufacturers outsource this part to the experts. Experts being companies that have years of experience doing backend design for ASICs of all kinds. So, the most notable one IMHO would be Innosilicon in China. They do tons of tape outs for many industries inside and outside of crypto. At this point you would tell them you want a 28NM chip or whatever transistor size you want and give them the RTL. They will validate again, and create the backend and PCB (if not already designed) and then essentially send off what is known as a GDS to a foundry.

Step4: The foundry will take the GDS and make what is known as the mask. The mask is pretty much what it sounds like. It is a piece of material they will lay on top of the Silicon Wafer and then after various processes including, doping, and photoetching, create the asics. This mask is very valuable and where most of the money in a tape out comes into play. With it, you can make that chip.

Step5: If you use a notable backend company to do your design, then generally there is no need for a so-called engineering batch. Most well-known companies will merge your RTL with their tried and true backend to make it a sure fire home run. If you do a engineering batch, it is because you are using an untested backend design where making a full scale production run may be a waste of money if you need to make tweaks to the backend. So, the chips are made, the PCB is made, now just assemble it, put a power supply in an enclosure, and sell it right? NOPE!

Step6: This is where some really important things come for global commerce. Firstly, if you do not have all the regulatory clearings from some big names in safety, some countries will block your equipment in customs. This is generally a must-do although there are some "creative" ways around this. Also, you need to test the chips yourself to make sure it performs as expected. Also, you need to develop your drivers for the machine. So if you are running a raspberry pi as a controller, you have to make a driver so CGminer can recognize it along with maybe making your own fork off cgminer. Then comes the task of making it user friendly with some sort of webGUI.

That is generally the process. Now as far as cost goes here is a break down.
NOTE: the larger the transistor size, the cheaper the manufacturing process. The prices below are for 28NM however if you go 20NM or 16NM these prices will surely go up ten-fold

RTL: 50,000-100,000 USD depending on if you do it or if you hire someone (generally just a salary)
Backend: 300,000 USD
PCB Design: 10,000USD
Mask Making: 1.1M USD
Chip Cost per chip:(greatly varies on many things)
things that affect chip price are:
RAM TYPE: ECC or Non-ECC
RAM AMOUNT
Number of Cores
Number of Layers
Generally speaking for a crypto chip the price per chip varies. For a SHA-256 chip there is generally no RAM in these chips. so these chips are cheaper. Now scrypt chips and others require alot of ram so they are more expensive. With that being said chips can range from $2-$15 each.

So you must pay for the number of batch1 chips

All in all the total cost comes to around 3M USD give or take..

Scenario 2: Buying Chips

Generally this can be a good idea because of cost savings, however, if will need to proper logistics in order to physically get those chips on a PCB and hook them up to a controller of some kind. If you have those avenues sorted already, it may be a good move. However, if not, it will quickly get hairy.

Hopefully this answered some questions

HI I want to do the same manufacter asic chips for bitcoin 16nm or 8NM 150GHS each i want to know steps and if u have business plan for this ;; what if that the models exist what s the cost for that please i need business plan of asic production