By Majeed Ahmad, contributing editor
In the late 2000s, when the PC industry was showing signs of slowing down, the iPhone erupted onto the mobile scene and reinvented the smartphone industry altogether. And the smartphone became a key driver for the semiconductor industry at a time of lumbering PC market growth.
The smartphone party went on for nearly a decade before it also began showing signs of a slowdown. Then one day, Intel announced that it was buying the FPGA powerhouse Altera. On December 28, 2015, almost everybody had been wondering about why Intel acquired Altera.
Ironically, at that time, the PC hardware king was mired in a battle for the riches of the smartphone, to which it had been dreadfully late. What industry watchers eventually found out can be summed up in two words: data center.
The data center had been rapidly evolving in an increasingly connected world while catering to new data sources such as smartphones, tablets, home appliances, wearable devices, and connected vehicles. Intel already had skin in the data center game with its Xeon CPUs, and the world’s largest chipmaker wanted to make its bets in the data center business even safer.
The logic of Intel’s acquisition seems self-evident now. The data center is the next growth driver for the chip industry after personal computers and smartphones. And it impacts almost every facet of semiconductor design.
Chips for data centers
Take, for instance, power electronics in which analog/mixed-signal design houses like Maxim Integrated and Vicor are creating low-power systems for data centers around initiatives such as the 48-V Direct-to-PoL architecture.
Then there is the emergence of high-speed serial links like PCIe 4.0 and active optical cable (AOC), which have been greatly influenced by the insatiable demand for data capacity in data center environments. And that will reshape design areas such as SerDes, Gigabit Ethernet, and interconnects.
Next, memory architectures are being redefined to meet data centers’ never-ending demand for speed and bandwidth. The DDR5 standard is around the corner and is going to double the bandwidth and density over DDR4. Moreover, DDR5 promises to deliver improved channel efficiency.
Also worthwhile to mention is the high bandwidth memory (HBM) version of DRAM architecture, which is significantly raising the bar on memory capacity and data throughput by providing a wider interface to SoC designs. A greater memory bandwidth and density closer to the CPU also brings significant efficiency benefits to server and networking systems in data centers.
The 3D XPoint non-volatile memory technology jointly developed by Intel and Micron is another example of how greater bandwidth and storage capacity are pushing the limits in the memory realm. Memory chips built around the 3D XPoint technology are expected to arrive later this year.
The GPU and FPGA chips represent another facet of how the data center is reinvigorating the semiconductor industry. Take GPUs, once used for scientific computing and high-end gaming: GPUs are now being seen as a key enabler of machine-learning functions carried out from data centers and cloud servers.
And FPGAs, armed with flexibility and reconfigurability, offer even more optimal solutions in the data center environment. They provide hardware acceleration for a wide array of applications while being paired with CPUs.
That brings us back to the question of why Intel acquired Altera. Intel CPUs paired with Altera’s FPGAs is akin to general-purpose computing coalesced with the application-specific hardware acceleration — a match made in data center heaven?
Advertisement
