New microprocessor architecture

"Torben "&#4294967295;gidius" Mogensen" <torbenm@app-0.diku.dk> wrote in message 
news:7zlkiafz9e.fsf@app-0.diku.dk...
> "Wilco Dijkstra" <Wilco_dot_Dijkstra@ntlworld.com> writes:
>
>> Incidentally, ARM2 is now over 20 years old, so the earliest patents
>> must be close to running out. Will anyone bother cloning an ARM2?
>> I think it's very unlikely - ARM2 was obsoleted by ARM7 in 1995.
>
> While ARM2 has been superseded by later processors, I don't think it
> is useless and not wortwhile to copy.  Granted, it can not compete
> with the later ARM processors, but it is smaller, so it could be
> useful for embedded or low-cost handheld designs (or something like
> the $100 laptop).  And there is a lot of software (from the Archimedes
> homecomputer) that can run on ARM2.
>
> I'm not saying people would rush to do it as soon as the patents run
> out, but I wouldn't rule the possibility out.

It would be relatively easy to make an ARM2 equivalent indeed. With
a modern process it would be as fast as an ARM7. Of course without
Thumb it would it would need 50% more flash, so ARM7 has the area
advantage. Then there is Cortex-M3...

A StrongARM equivalent would be a lot more interesting, but it has
to avoid the more recent MMU and cache patents. Even so, a
64MByte address space is pretty limiting for a $100 laptop...

Also it's not obvious to me that designing your own ARM would be
cheaper than licensing (the development cost of a core is spread
over many licensees). It only makes sense for very high volumes as
you avoid the royalties. I don't rule it out entirely - I can see it done
as a free FPGA core for example.

Wilco 

"Wilco Dijkstra" <Wilco_dot_Dijkstra@ntlworld.com> wrote in message 
news:2l%Gh.44598$3a3.43658@newsfe6-win.ntli.net...
>
> "rickman" <gnuarm@gmail.com> wrote in message 
> news:1173118005.268771.265510@8g2000cwh.googlegroups.com...
<snip>
>
>>The only real
>> barrier to potential customers developing or using open source ARM
>> designs is the patent ARM has on an essential circuit I am told. 
>> When
>> that runs out in another 10 or so years, ARM7 will be freely copied 
>> if
>> anyone still has interest in it.  Maybe that is why they have come 
>> out
>> with the Cortex M3, to start a new patent clock ticking?
>
> Not really. ARM7 is 12 years old and already at the end of its useful
> life, so it was becoming urgent to design something better. It's 
> always
> better to obsolete your own products rather than let a competitor do 
> it!

I'm not sure that is completely true - there's plenty of mileage in the 
ARM7 yet and I'll bet there will still be plenty of demand for it for 
the remaining life of the patent and beyond.

Sure there are newer cores taking off but they are linked to phones and 
PDAs and the question of longevity of those parts will keep designers in 
the security of the established ARM7 lines. 

On Mar 5, 11:45 pm, Grant Edwards <gra...@visi.com> wrote:
> On 2007-03-06, Jim Granville <no.s...@designtools.maps.co.nz> wrote:
>
> I'm working on a product line that just migrated from ARM7 to
> NIOS2.  Nobody cares about the CPU core.  We're using the same
> OS and same compiler.  Changing CPU cores was trivial -- and it
> even involved going from big-endian to little-endian.
>
> Changing Ethernet controllers and various other peripherals was
> the painful part. ;)
>
> It's the peripherals where all the work is -- the CPU just
> doesn't matter.

I'm not trying to sound stupid (but we all do what we do best), but if
the core "does not matter" then why did you change?  I guess the
answer is you are moving to an FPGA or possibly even an ASIC?

I think your statement is a bit overbroad.  Of course the CPU core
does matter.  But the requirements on the core are thinks like speed,
support, etc., that many CPUs will meet.  Then the choice becomes
about selection in the other areas to narrow it down to a chip.  At
the top of this list are the peripherals which is what you are saying
is the main consideration.  Certainly it depends on your application.
You just said you were changing the core in spite of the fact that you
now have to change all your peripherals!

I am pretty sure I have seen 8051 chips with Ethernet.  But I doubt
that you considered them as useful devices in your initial design.
You wanted the higher MIPS of the ARM and the greater flexibility in
finding parts from multiple vendors with a wide selection of
peripherals.

BTW, when you say you are using the same OS and compiler, do you mean
the exact same tools or new tools from the same vendors?  I don't know
of any tool providers that bundle NIOS2 and ARM in the same package.
Just curious, which tools are you using?

Am I missing the point?

On Mar 6, 5:16 am, "Wilco Dijkstra" <Wilco_dot_Dijks...@ntlworld.com>
wrote:
> I'm saying that you don't need a patent immediately. Only when you've
> got a product that works (and we agree that takes a lot more than just
> having a patent) and are ready to sell you need patents.
>
> However a patent means the invention is published, so anyone (say
> in a country where you didn't patent) could copy it easily. World-wide
> patents are very expensive, so most only patent in the US and UK.
> Patents are very useful when you're selling things that can be easily
> copied or reverse engineered, but they are a double-edged sword.

And while you work on your marketing plan someone else applies for the
same patent (which is not unusual) you are left out in the cold.  I
believe it is pretty universal advice to apply early for patents.


> Yes, without patents the IP would be unprotected. I think you agree...

Just the opposite.  Most IP can not be protected by patents.  Only
inventions can be patented.  That is why all the IP companies use
encrypted libraries and licences to protect their IP.


> Indeed. My guess of what would happen is that everybody creates
> incompatible extensions and there is less incentive to invest into
> innovative designs as they are easily cloned. We'd still use ARM2's...

Yes, that is likely true, except that I don't see that the ARM2 was so
wildly successful.  I would say it was the ARM7/9 that gave ARM the
big boost that is making the ARM a commodity device.



> Intel and AMD are a duopoly, and yes it's great there is fierce competition
> between them (thanks for reminding me), but that's not the subject of
> discussion. My point is that the cost of making an x86 clone is far higher
> than making an ARM clone, so protection is more important for ARM.
>
> Fabs are an important difference: if you have a fab you sell chips to
> end users, not IP to companies with fabs. Without patents you can still
> sell chips (end users can't make chips themselves), but it becomes a
> lot harder to sell IP to a fab company (they can make the IP themselves
> and avoid royalties). Selling IP is hard with patents - and impossible
> without.

So Xilinx does not sell chips?  What about the other hundreds of
fabless semi companies?  Your facts are wrong and so your conclusions
are wrong.

BTW, if patents are so much more important to ARM than Intel, why does
Intel have so many patents?  Not only on their chips, but on many
aspects of making chips!!!  Your logic just does not wash.


> They would say that, wouldn't they? If you read ARM's latest financial
> statements you can see that licensing of ARM7 is clearly declining fast
> (under 5% of total license revenue in 2006, down from 18% in 2002).
> That doesn't imply it won't stay popular for the next few years, but ARM9
> and Cortex-M3 are appearing fast.

Yes, they say it and they do it!  Atmel has some half dozen new ARM
chips coming out this quarter. Philips is introducing new ARM chips
continuously.  ST Micro has a whole new family of ARM7 chips this
quarter.  The list goes on....

I don't know your licensing numbers, but I know to license an ARM7 you
have to pay a large, up front fee along with a per chip fee.  You only
pay the up front fee once, so it is natural for the ARM7 license fees
to drop as a product matures.  But your numbers don't actually say the
ARM7 license fees are dropping.  They say they are a smaller part of
the total company revenue.


> Yes, I'd like to see ARM7 and Cortex-M3 on the same process

The odd thing is that when I spoke with LM about it they did not say
they were going to a 180 nm process like everyone else.  They plan to
intro a new family on the same 250 nm process, but with lower power
and then intro a new family in 130 nm late in the year.  I have not
been able to get any new info on this so they may be slipping on their
schedule.


> The patents are there to protect the investment and clearly don't
> lock out competition - they just stop cloners.

You didn't read the entire sentence.  That is what I mean by lockout.
They stop the competitors from duplicating the product in terms of the
ISA.  No one can produce an ARM7 without violating a patent.

On Mar 5, 10:59 pm, "Michael Brown" <s...@signature.below> wrote:
> Unsurprisingly, AMD reacted with x86-64, and so far IA64 is showing no signs
> of an imminent takeover ...


A true master of understatement!

On Mar 6, 7:30 am, "Wilco Dijkstra" <Wilco_dot_Dijks...@ntlworld.com>
wrote:
> A StrongARM equivalent would be a lot more interesting, but it has
> to avoid the more recent MMU and cache patents. Even so, a
> 64MByte address space is pretty limiting for a $100 laptop...

Sounds like a Bill Gates statement...  "Nobody needs more than 640KB
of memory". <G>


> Also it's not obvious to me that designing your own ARM would be
> cheaper than licensing (the development cost of a core is spread
> over many licensees). It only makes sense for very high volumes as
> you avoid the royalties. I don't rule it out entirely - I can see it done
> as a free FPGA core for example.

The cost of the license is not connected to the cost of the
development other than providing enough money to keep the company
afloat.  The will charge as much as the market will bear which means
they can extract a significant percentage of the development cost from
each licensee.  In the end they get many more times their costs back
in fees.

In terms of a company duplicating the ARM ISA, I think it may well be
cheaper to do that than to license the core.  By copying you avoid all
the costs of inventing the ISA, working with tool vendors (or even
paying them) for tool development and all the other costs that ARM
has.  But you can only get this advantage if you copy the ISA of an
existing MCU like the ARM.

Of course you could go with an open source CPU like the one from
Lattice.  No license fees, no royalties and you have to do all the
real work yourself!

On 2007-03-06, rickman <gnuarm@gmail.com> wrote:
> On Mar 5, 11:45 pm, Grant Edwards <gra...@visi.com> wrote:
>> On 2007-03-06, Jim Granville <no.s...@designtools.maps.co.nz> wrote:
>>
>> I'm working on a product line that just migrated from ARM7 to
>> NIOS2.  Nobody cares about the CPU core.  We're using the same
>> OS and same compiler.  Changing CPU cores was trivial -- and it
>> even involved going from big-endian to little-endian.
>>
>> Changing Ethernet controllers and various other peripherals was
>> the painful part. ;)
>>
>> It's the peripherals where all the work is -- the CPU just
>> doesn't matter.
>
> I'm not trying to sound stupid (but we all do what we do best), but if
> the core "does not matter" then why did you change?  I guess the
> answer is you are moving to an FPGA or possibly even an ASIC?

An FPGA into which we can shovel the core as well as
peripherals that used to be off-chip in a fairly large ASIC.

> I think your statement is a bit overbroad.  Of course the CPU
> core does matter.  But the requirements on the core are thinks
> like speed, support, etc., that many CPUs will meet.  Then the
> choice becomes about selection in the other areas to narrow it
> down to a chip.  At the top of this list are the peripherals
> which is what you are saying is the main consideration.
> Certainly it depends on your application. You just said you
> were changing the core in spite of the fact that you now have
> to change all your peripherals!

We're going to a one-chip solution instead of a 2-4 chip
solution.  Spending NRE to reduce RE.  The ARM7 part was also
being discontinued, so we were going to have to change anyway.
We've since been told by the ARM7 vendor that they've changed
their mind and are not disconinuing the part.

> I am pretty sure I have seen 8051 chips with Ethernet.  But I
> doubt that you considered them as useful devices in your
> initial design. You wanted the higher MIPS of the ARM and the
> greater flexibility in finding parts from multiple vendors
> with a wide selection of peripherals.

Actually the fact that it was ARM didn't matter.  We were
looking for a peripheral set.  Pretty much any 32-bit processor
running at 20-50 MHz would be fine (ARM, PPC, SPARC, etc.).

> BTW, when you say you are using the same OS and compiler, do
> you mean the exact same tools or new tools from the same
> vendors? I don't know of any tool providers that bundle NIOS2
> and ARM in the same package. Just curious, which tools are you
> using?

GCC and eCos.  GCC and eCos will build for at least a dozen
target architectures.  eCos for both targets is built from the
same source tree.  Right now we've got two different gcc source
trees, but that's just because the ARM toolchain is a couple
years old and the NIOS one is a little newer.

Once the device drivers for the peripherals were working, it
took about two days to intially port the apps. It'll probably
another week or two of testing and fixing minor problems).

> Am I missing the point?

I don't think so.  In my experience, what matters is the
peripheral set and the availability of development tools. Given
an appropriate peripheral set and OS/toolchain, we could just
as easily be using any of four or five other 32-bit CPU cores.

-- 
Grant Edwards                   grante             Yow!  These PRESERVES
                                  at               should be FORCE-FED to
                               visi.com            PENTAGON OFFICIALS!!

 Its impossible to patent what no one needs .

 CPU's are fast to change , even faster

 is software .   Software change forces

 even better , more competitive H/W .

  CPU's have been highly regulated , thus

 evolution was slow , thank to Intel crony .

 But now the consumer is getting better ,

 for the wide competition of ARM cpu .

  There are no shortages of RAM ,ROM

 and inspite of C/C++ and linux and M$

 those who need good software are using

 it now . Dont expect them to advert' it ,

nor try to sell it to the public , it would

 work against their interest .

   Hardware makers are hoping their software

 will "lock" the consumer into buying the

 pdt .  If the software runs on another

 box , they may only sell one box , softwa

 is carried to their competitors box ,

 and they go bankrupt .

   As competition gets stronger , ppl give

 away free software ...

  Patents and copyrights become a joke .

 M$ only had a copyright as long as Intel

 had control on CPU's ..

  Linux will be history soon ..

   I am the worlds fastest systems programmer .
 I will "borrow" Nintendo DS  BIOS and then
 rewrite it and slim it down and put a perfectly
 structured OpSys on top of that and ..

   Give it free .

 Because it nix's all the difficulties of programming ,
 it will be used by all .
  No files nor folders , kernel organizes everything
 for you , and to boot has total control over
 every thread and object , thus impossible to
  virus , worm ...

"rickman" <gnuarm@gmail.com> wrote in message 
news:1173190734.522830.23610@j27g2000cwj.googlegroups.com...

> In terms of a company duplicating the ARM ISA, I think it may well be
> cheaper to do that than to license the core.  By copying you avoid all
> the costs of inventing the ISA, working with tool vendors (or even
> paying them) for tool development and all the other costs that ARM
> has.  But you can only get this advantage if you copy the ISA of an
> existing MCU like the ARM.

I can't see how it could be cheaper. An average ARM license is around
$2.5m, ARM7 should be below $1m. Could you clone an ARM for less?

Wilco 

Grant Edwards wrote:
> We're going to a one-chip solution instead of a 2-4 chip
> solution.  

I can see the motivation, but is it really a one chip solution ?
Most FPGA designs are 3 chips :
FPGA device, Loader Memory, and Code memory.
Yes, Loader memory is often small, and in
the newest Xilinx devices it's in the package, but
the code memory is more of an issue, as that's wider.

Or did you mean 'One complex chip', or 'one chip to learn' ?
- as one could say the support memory devices are largely invisible
from a devlopment viewpoint.


> Spending NRE to reduce RE.  The ARM7 part was also
> being discontinued, so we were going to have to change anyway.
> We've since been told by the ARM7 vendor that they've changed
> their mind and are not disconinuing the part.

Who was the ARM vendor ?

I've seen that flip-flop before, but usually the continued part also 
moves into 'maint pricing' columns.

-jg