Why did Microsemi buy Actel?

[John Blyler]

Three weeks ago, Intel announced the mid-year 2011 availability of the
first programmable embedded ATOM SoC – codenamed Stellarton – based on
Altera’s FPGA technology. Earlier this year, Xilinx announced a
partnership with ARM, the current de facto leader in embedded mobile
systems. Both of these announcements were processor-centric, i.e., an
embedded processor was tightly couple to an FPGA. (See “Intel Teams Up
with Altera”)

This is not the case with today’s announcement of Microsemi’s
acquisition of FPGA tool vendor Actel. Rather than a marriage of
processors with FPGAs, this announcement represents a union of analog-
mixed signal (AMS) and RF/Wireless chips with FPGAs. Why the
difference?

http://www.chipdesignmag.com/blyler/2010/10/04/why-did-microsemi-buy-actel/

 

[rickman]

On Oct 4, 6:01 pm, John Blyler <john.bly...@gmail.com> wrote:

Three weeks ago, Intel announced the mid-year 2011 availability of the
first programmable embedded ATOM SoC – codenamed Stellarton – based on
Altera’s FPGA technology. Earlier this year, Xilinx announced a
partnership with ARM, the current de facto leader in embedded mobile
systems. Both of these announcements were processor-centric, i.e., an
embedded processor was tightly couple to an FPGA. (See “Intel Teams Up
with Altera”)

This is not the case with today’s announcement of Microsemi’s
acquisition of FPGA tool vendor Actel. Rather than a marriage of
processors with FPGAs, this announcement represents a union of analog-
mixed signal (AMS) and RF/Wireless chips with FPGAs. Why the
difference?

http://www.chipdesignmag.com/blyler/2010/10/04/why-did-microsemi-buy-...

Microsemi may have a leg up with a few Wifi like chips which could be
integrated with the Actel SOC devices. I am sure there are markets
for many combinations of these devices. I wonder where the next
killer app will come from.

I had not heard of the Intel/Altera union. That appears to be
oriented to creating a device that can be responsive to rapidly
changing I/O requires of the market. It can also allow a single
hardware device to address many smaller markets rather than requiring
a separate chip for each. But what exactly are those markets?
Currently the Atom processor is a bit "heavy" for truly portable apps
such as cell phones and devices of similar size. I expect it is even
too much for PDAs.

The Xilinx/ARM announcement may end up being much more interesting,
potentially combining a range of processors and a range of FPGAs on a
single die. There is a virtually unlimited range of markets for such
devices. Being on a single die will greatly lower the cost and
size.

But all of this is far out and we can only wait for more details. I
expect chips will not be in our hands for years to come.

Rick

 

[Eric Smith]

rickman wrote:

The Xilinx/ARM announcement may end up being much more interesting,

I'm not sure why to expect it to be any more interesting than the
Altera/ARM stuff was, or for that matter, the Xilinx/IBM stuff. As
in, not very interesting at all.

Eric

 

[rickman]

On Oct 5, 1:36 am, Eric Smith <space...@gmail.com> wrote:

rickman wrote:
The Xilinx/ARM announcement may end up being much more interesting,

I'm not sure why to expect it to be any more interesting than the
Altera/ARM stuff was, or for that matter, the Xilinx/IBM stuff.  As
in, not very interesting at all.

Eric

Why was the Altera/ARM stuff not interesting? It failed because it
was so expensive, in my opinion. I expect we'll never find out from
Altera. The Actel Fusion part is also a bit pricey. You forgot to
mention the Atmel FPSLIC part (Atmel FPGA/AVR). I'm not sure if that
is still around, but it has become inconsequential because of the
price and the lack of advancement of the development tools.

The Xilinx/ARM stuff might be interesting if they actually tap the
full potential of providing a range of products to match the range of
CPUs that are available from ARM. If Xilinx just repeats what they
did with the PPC and just makes one limited line of devices then I
don't see that being especially successful either.

Actually, I shouldn't say that. Knowing Xilinx, I expect they will
identify the markets with the most potential and design a small number
of FPGA/ARM devices tailored to those apps. One might be cell
phones. That is a potentially HUGE market and SiBlue is currently
trying to break into that area. If Xilinx combines an ARM with a low
end FPGA like the SiBlue parts, they could beat them at their own
game. Or perhaps a larger FPGA (if it is cheap enough and low power
enough) could replace some of the custom logic in a cell phone? But
mostly, Xilinx understands their core market, network comms. For them
a large FPGA combined with one or more high end ARM cores (a la Virtex-
II Pro) is what we can expect.

If you were combining an FPGA with a CPU in a market where a number of
companies were doing that, how would you differentiate your product?

Rick

 

[dalai lamah]

Un bel giorno rickman digitň:

I had not heard of the Intel/Altera union. That appears to be
oriented to creating a device that can be responsive to rapidly
changing I/O requires of the market.

I think it's more likely that they are looking for a way to crunch numbers
faster and close the gap with GPUs:

http://blogs.nvidia.com/ntersect/2010/06/gpus-are-only-up-to-14-times-faster-than-cpus-says-intel.html

--
emboliaschizoide.splinder.com

 

[Ulf Samuelsson]

rickman skrev:

On Oct 5, 1:36 am, Eric Smith <space...@gmail.com> wrote:
rickman wrote:
The Xilinx/ARM announcement may end up being much more interesting,
I'm not sure why to expect it to be any more interesting than the
Altera/ARM stuff was, or for that matter, the Xilinx/IBM stuff. As
in, not very interesting at all.

Eric

Why was the Altera/ARM stuff not interesting? It failed because it
was so expensive, in my opinion. I expect we'll never find out from
Altera. The Actel Fusion part is also a bit pricey. You forgot to
mention the Atmel FPSLIC part (Atmel FPGA/AVR). I'm not sure if that
is still around, but it has become inconsequential because of the
price and the lack of advancement of the development tools.

The FPSLIC is still in production, but since no shrink has been made,
there is no long term migration path.
Many of people working with the FPSLIC left and started at Actel,
so mayube the Fusion stuff is really FPSLIC-2...

It certainly addresses many of the concerns of the FPSLIC.
1) No Analog
2) No Flash
3) VERY limited code space (32kB max)

I have seen what appears to be the perfect application for the Fusion,
and it can be solved at 50% of the Fusion cost using 2 micros.

If you want mixed signal CPU/FPGA devices,
then you have to use the same processes as the standard micros.
This makes the part too expensive.

If you want to go for low cost, you go real aggressive pure digital CMOS
and have to use external memory, or large SRAMs.
Most companies I see that is looking for 32 bit MCUs require
256 kB - 1 MB code memory, which has to be implemented
as SRAM on the chip, if you do not have a flash capable process.

I think that you either have to go for application processors
like ARM9 or above, or limit the FPGA to a small part of the chip
for the cost to be acceptable.

One problem that you face when you design such a chip, is that you need
to make a family, for people to accept.
You have memory size in one dimension and FPGA size in another dimension.

If you have 4 different memory sizes and 3 different FPGA sizes,
you will have to create 12 different masksets.
This will be very expensive in 28 nm...

The Xilinx/ARM stuff might be interesting if they actually tap the
full potential of providing a range of products to match the range of
CPUs that are available from ARM. If Xilinx just repeats what they
did with the PPC and just makes one limited line of devices then I
don't see that being especially successful either.

Actually, I shouldn't say that. Knowing Xilinx, I expect they will
identify the markets with the most potential and design a small number
of FPGA/ARM devices tailored to those apps. One might be cell
phones. That is a potentially HUGE market and SiBlue is currently
trying to break into that area. If Xilinx combines an ARM with a low
end FPGA like the SiBlue parts, they could beat them at their own
game. Or perhaps a larger FPGA (if it is cheap enough and low power
enough) could replace some of the custom logic in a cell phone? But
mostly, Xilinx understands their core market, network comms. For them
a large FPGA combined with one or more high end ARM cores (a la Virtex-
II Pro) is what we can expect.

If you were combining an FPGA with a CPU in a market where a number of
companies were doing that, how would you differentiate your product?

Rick

Best Regards
Ulf Samuelsson