stuff that makes you go hmmm. This guy may have read the thought provoking essay in 2 parts on
the evolution of the sensor hub written by Paul McWilliams? No way to know who he is talking about in the bolded section, but very thought provoking!
Intel: The Magic Of FPGAs
Jun. 2, 2015 2:17 PM ET | About: Intel Corporation (INTC), Includes: AAPL, ALTR, XLNX
Disclosure: The author is long INTC. (More…)
Summary
•Intel is offering a ton of money for Altera.
•What is an FPGA.
•Is Intel + Altera a 1 + 1 = 3 kind of deal.
FPGAs are the ultimate Moore’s Law component.
FPGAs (Field Programmable Gate Arrays) came on the scene from Xilinx (NASDAQ:XLNX) in 1984. They were slow, low density, expensive, power hogs, but easy to configure for a small custom logic design. The first couple of generations were pretty useless except for very specialized applications. We had a term for the usage case for FPGAs; “Protoduction”. That meant that FPGAs could be used in certain prototype and low volume production applications. If an application was successful and went to high volume, the FPGA was replaced with and ASIC (Application Specific Integrated Circuit) at one tenth the price.
The state of the semiconductor technology back then was two micron and logic layout and design was cheap. The promise for FPGAs was that as the semiconductor technology advanced and logic design for ASICs became more expensive, FPGAs would find a more mainstream market than “Protoduction”.
As of 2014, the FPGA market is somewhere around $4.5 billion, primarily divided between Altera (NASDAQ:ALTR) and Xilinx. The largest use markets for FPGAs have in communication infrastructure and industrial applications.
FPGAs are funny little critters. They come from the factory as blank arrays of “gates”, the building blocks of all things digital. Think of them as blank black boards. The logic designer, using very sophisticated CAD (Computer Aided Design) tools, designs the logic that will occupy this blank piece of silicon. When that design is completed, it is loaded into the FPGA and, like magic, the blank piece of silicon becomes a digital electronic function from something simple to something intensely complex. Also like magic, if the power is turned off, that FPGA goes back to a blank piece of silicon ready for another logic function to be loaded. So, a clever designer can use the same piece of silicon for many different logic designs, as long as the functions don’t have to operate at the same time. Dr. Moore must giggle when thinking about FPGAs…the same set of transistors used over and over again for different hardware function. This is Moore’s Law 2.0.
In the 30 years since FPGAs appeared, the cost (area) of transistors has decreased by a factor of 20,000 (2,000 nm to 14nm). At the same time, the cost of designing logic ASIC devices has increased geometrically to the point that a complex ASIC can cost hundreds of millions of dollars to design and get ready for production in a bleeding edge semiconductor technology node.
Recently, these cost curves have caused a tipping point to appear that makes it easier, quicker and cheaper to use FPGAs in all but the very highest volume custom chip application. While this has happened, the FPGA has retained the absolutely unique ability to change function on the fly, thus multiplying its value and functionality.
Intel (NASDAQ:INTC) is in the process of buying FPGA producer Altera because many data center designers and operators have recognized the value of FPGAs used in conjunction with high performance server CPU chips to enhance the throughput and reduce the power requirement of data centers. There are immensely complex algorithms that can now be performed by the hard logic of an FPGA at much higher speed and much lower power than could be done by the server chip alone. These functions can also change, if necessary, by simply sending the new configuration file to the FPGAs.
Outside of the data center, imagine a small FPGA embedded in a smartphone application processor and think about the potential for hardware logic based apps vs. software apps. Think about logic designs that are unique to a particular smartphone manufacturer.
What if Apple (NASDAQ:AAPL) whispered to silicon suppliers that the next A chip supplier would not be the guy with the smallest node, but the guy who could supply a chip with a section of programmable logic to be used by Apple for the “next big thing”?
Intel might have paid too much for Altera, or they might have paid a reasonable price for company that has just hit a tipping point that is not recognized by the head-scratching analysts yet. The deal is slightly accretive to Intel’s earnings and may get much more so over time. One thing for certain, Intel has taken Altera out of the realm of fabless semiconductors back into the world of IDM (Integrated Device Manufacturer) with the attendant significantly lower production costs. Xilinx should be very, very afraid. There is no equivalent acquirer for Xilinx that has a lock on the server business like Intel has.
The acquisition of Altera by Intel opens opportunities that didn’t exist for either company separately.
In this piece, I haven’t mentioned the $20 billion 3D NAND based storage opportunity for Intel.
