Do you feel a ripple in the force? As you are doubtless aware, the hot news this week is that microprocessor behemoth Intel is purchasing programmable logic giant Altera (see Intel to Buy Altera for $16.7B by yours truly, Intel, Altera: Math in Questionby Rick Merritt, and 2015 Mergers, acquisitions and change in the global electronics industry by Steve Taranovich).
This has sparked all sorts of discussions and questions. One that interests me in particular is: “How will this acquisition affect embedded space?”
But before we go there, let's briefly peer into the mists of time… Intel was founded in 1968 by Gordon Moore (of “Moore's Law” fame), a chemist, and Robert Noyce, a physicist and co-inventor of the integrated circuit. Intel actually started out as a memory company; its first product was a 64-bit static SRAM in 1969. Even after Intel created the first commercially available microprocessor — the 4004 — in 1971, followed by a raft of increasingly sophisticated descendants, its business continued to be dominated by DRAM chips into the early 1980s.
Simple programmable logic devices (SPLDs) first appeared on the scene in the early 1970s. They started with PROMs in 1970, followed by PLAs in 1975, PALs in the late 1970s, and GALs in the early 1980s. The early 1980s also saw the emergence of more sophisticated components called complex programmable logic devices (CPLDs). A big leap forward occurred when newly-formed Altera Corporation introduced a CPLD based on a combination of the CMOS and EPROM technologies. Meanwhile, in 1985, recently-formed Xilinx presented the world's first FPGA to the market.
It probably won’t surprise you to discover that Intel had its own Programmable Logic Device (PLD) Business Unit, and that they offered families of programmable “iPLD” components ranging from SPLDs and EPLDs to FPGAs. The ironic thing here is that, in 1994, Altera increased its share of the programmable logic market to 20 percent by purchasing Intel's PLD business for ~$50 million in stock and cash.
But we digress… These days, we tend to think of Intel as predominantly being in the personal computer and server spaces. As I discussed in my EE Times column, Intel's processors are really good at performing decision-making operations and implementing control functions, but they are relatively inefficient when it comes to many algorithmic data processing tasks. By comparison, FPGAs can be used to perform computations in a massively parallel fashion, which makes them ideal for compute-intensive applications and tasks like communications infrastructure, machine vision, and big data processing.
Many servers already employ a mix of CPUs and FPGAs to implement high-performance computing (HPC) solutions. With Intel's fab and advanced packaging technologies, we can expect to see state-of-the-art processor and FPGA dice in the same package, and also processor and FPGA cores on the same die.
But what does all this have to do with embedded computing? Well, when you say embedded systems, many people tend to think of things like the Internet of Things (IoT), and — from there — they are inclined to think of incredibly small, low-power sensor and processing nodes “on the edge.” But “embedded space” covers much more than the IoT, and there's a tremendous market for embedded systems that can offer extreme processing performance.
Consider Intel's arch adversary AMD, which is its only significant rival in the central processor (CPU) market for (x86 based) personal computers. In recent years, Intel has taken the high-ground in the personal computer and server markets, but AMD is making significant thrusts into a variety of high-performance embedded markets. If AMD sees a market opportunity, then you can bet that Intel is interested also.
Chips that mix CPU and FPGA functionality on the same die offer all sorts of interesting possibilities. We should also remind ourselves that Intel isn’t totally tied to the highest performing and power-guzzling processors. Similarly, Altera does more than make the biggest and hairiest FPGAs. Consider Intel's Atom processors, which deliver energy-efficient performance to power devices ranging from smartphones and tablets, to intelligent cars, to cutting-edge healthcare devices, to microservers for the cloud, and … the list goes on.
Of particular interest here is the fact that Intel has already paired an Atom-series processor with an Altera FPGA in a single package. One limitation to this early incarnation is that all communications between the Atom processor and the FPGA have to funnel through a 1×1 PCIe channel linking the two die, thereby forming a bottleneck that makes offloading some tasks from the CPU to the FPGA impractical. Now that Intel has acquired Altera, we might start to see more innovative implementations in which the processor and FPGA dice are linked using silicon bridges that could potentially offer tens of thousands of inter-die connections.
To be honest, my mind is reeling with various possibilities — new ideas keep on popping into my head. How about you? What do you think about all of this? How do you think this acquisition might play out in embedded space (where no one can hear you scream)?