That's a really long graduation project. I'd say it seems unreasonable effort for a final year project. Something of this magnitude (2 years worth of work) looks more like a masters degree.



So, just some of the patents, then? You are aware of the complications of cloning a fully patented work, aren't you?
Let's say you write a processor from scratch that executes the ARMv8 ISA. Well, it is a violation of licensed IP. You will never be allowed to share your work, publish it, or even make it available here at OpenCores, for example. You'd need to keep it to your personal amusement.



Yes, looks like you need someone to explain it more to you. Linus never stole a single line of code from Microsoft Empire.
Linux is written from scratch, using solid licensing to back it up. The entire work of the Linux Foundation is protected under Copyright Law, and is licensed under a GPL license.

If Linus did something like you propose (copy an IP that is protected and violating its licensing terms), Linux would never be, and Linus Torvalds would never be able to topple Microsoft, because he would be sued to his bones and be forced to withdraw every bit of the work from public reach.

You can only share something that is YOURS in the first place, i.e., something that nobody can claim property but you. Then you may decide to license it with a GPL, or even a BSD license, and give it to the world.



I do not want to warn you against anything. This is a Forum, and you invited all passers by to talk. I am just pointing out the inconsistencies in your plan.

Copying *ANY* processor core with the complexity of an ARMv8 class processor is a gargantuan task, that would be daunting to a team of very experienced processor designers, and requires an army of backend and verification engineers, just to check if the design works. Doing it from scratch, from the ISA and specs, just compounds that difficulty. Companies like Texas Instruments, Marvell, Apple and Samsung pay licenses of hundreds of million dollars to acquire rights to design their own cores based on ARM's ISA, and they have entire buildings packed with the best paid processor engineers in the world. These teams may find 100 weeks dismally short to do that.

You referred to the ARM Cortex-A7 as a "simple" core, to be done with as a stage during the project. That sounds preposterous. I really cannot make any judgement on your abilities as a processor designer, and you can be a gifted hardware talent, but the fact that you are at least 2 years from graduation, and the fact that you refer to cloning a Cortex-A7 as an ultra simple task, do not talk well on your behalf.

So, please keep us posted when you do it.

- Jonny

I agree with everyone here, you are cloning a much tested architecture and I am building one complete with a fpu,registers, a new instruction set etc. and have to work around 14 hours a day. I just finished my paper design on FPU in 28 hours of work so you can gauge how complex would reading and emulating a "cortex" processor would be. You will have to write all the drivers,the co-processor interface,the task scheduler,the interrupt handler and even more you will have to make them compatible with their design or your product's market value is zero (that is if you manage to somehow sidestep the legal process).

@Manili, sorry for my duplicate post. Something seems to have gotten awry with the Forum interface.

I can think of several suggestions for you:

- A multicore interconnect for hardware scheduling on processor clusters
This is a very current problem, that can be tackled by VLSI. Achieving close to linear scalability on large arrays of multicore processor clusters has always been a very difficult task. If you can prove that a hardware scheduler can scale on a multicore processor, that would be a very good contribution. You can even port Linux to your system, using your hardware scheduler, and show how it scales and why.
Although this is not seminal work, it is a valuable field of research.

- Ultra low power processor design
Choose a generic processor (even an existing one), and implement it with ultra low power design techniques. For example, implement all subsystems (bus, memory interface, FPU, integer units, cache units, control) as independent units that can be clocked off and have power removed, and can be powered up in minimum time. If you show a working processor that can take advantage of conservative algorithm execution, and show low energy profiles based on a Power-Delay Product instruction scheduling, for example, where you turn off and on subsystems based on downstream needs of the instruction flow, that is significant even for a very simple processor architecture. Conservative algorithms have high data locality, and will not access cache or external memory during the core processing time, allowing the bus and memory interface to work less, or even be turned off during peak processing. You can even research adiabatic circuitry for some processor blocks.

- A simple, but complete, processor design that shows end-to-end implementation skills
For a graduation project, this is a significant task. Taking the 6502 suggestion someone made on this thread, it can be really worthy.
You can write a 6502 from scratch, and transform the PAGE0 addresses into CPU registers, or zero-access time on-chip SRAM registers, for example. You can design from scratch a RISC engine that executes the 6502 instructions, and provide a prefetch/decode/issue instruction unit that translates canonical 6502 instructions into your RISC engine instructions, and then run the processor in single-cycle. The fetch can be made in wide words, for example fetching in 32 bits, and executing in single-cycle. If you do not like the 6502, take any other 8bitter with this approach.

There is a plethora of subjects you can take that can be done by a single person, and can show design capability, ingenuity, research, and new approaches. Actually, depending on your design, a single circuit element can make your PhD and a valuable patent. Massively huge boilerplate code may not be as significant as a new approach to a simple problem.

- Jonny

@Jonny
First of all there was a typo in my first post : I wrote "simple cores like Cortex-A7" but I meant "simpler (than Cortex-A57) cores like Cortex-A7" ! So I repeat that was a typo.

I just read your advices and they are worthless ! Thank you ! Is it possible for me to contact you about the advices ( e.g. via Email ) ?

(M. A. nili sent me private mail, bsically saying that my advice is worthless, and asking for help on suggesting a good idea.)

Hello Mohammad,

I am a bit puzzled.
You said my advice is worthless, but still, ask for help. Please make your mind.

Here is my uncut take on your situation:

You opened a thread for comments on a project for an ARM Clone. That is plainly not a good idea, for all the reasons that I mentioned in the Forum. You are not going to make a dime on patented work that you infringe.

Having said that, there are certainly numerous ideas for a Hardware Startup. I am designing in my company SoCs for use in smart metering and high voltage smart front-ends using 350nm 700V CMOS and 180nm standard CMOS with security and integrated failsafe, for example.

If you need to work on a commercially viable VLSI idea, start by studying the ecosystem, and decide whether you will design chips for applications, or design technology for chipmakers. Your first idea of designing a complete processor IP does not seem to be very wise, especially if you will compete against ARM.

Something that works is to design IP to be integrated in SoCs. Peripheral interfaces, logic that can be integrated into operating systems and bare metal systems to overtake software functionality, massive parallel schedulers, graphics and vision systems.

I have the strange impression that nothing I can say will be considered "worthy" by you. However, instead of taking your offensive remark as such, I am trying to interpret your comment as lost in cultural translation. If, on the other hand, your intention was really to offend, all I can say is that the world is much more ruthless than University. Use your time to learn and become very good at whatever you choose to do.

- Jonny

[P.S.: have you looked into OpenRisc project? It is an OpenSource project with very talented and serious maintainers, and some commercial applications. You could do work related to OpenRisc in your graduation/MS project. Worthless still? ]