The perception among Intel bears and many analysts is that Intel is in trouble for: no growth, can't innovate, PC is dead, missed mobile, is too arthritic to survive, etc., etc.
Reality, however, is a horse of an entirely different hue.
Since the 90nm technology node of the early 2000s, Intel alone has been the technology pathfinder that allowed Moore's Law to continue to shrink the feature size and improve the performance and lower the costs of integrated circuits.
Gate first/ gate last:
Virtually the entire industry decided to go "gate first" processing at 45nm. Intel (NASDAQ:INTC) alone went "gate last". TSMC (NYSE:TSM) also went gate last only after Chairman Morris Chang dictated that TSMC follow Intel. There were many reasons at the time that the IBM (NYSE:IBM) led Common Platform Consortium felt that gate first was superior, but the real, behind the scenes, reason for the gate first decision was that IBM felt that Moore's Law stopped at 45nm, so there would be no need for gate last. Intel, on the other hand, knew there were more steps to Moore's Law all the way to TriGate, and a TriGate transistor is impossible using gate first. Intel had the leading insight in this major decision.
At 90nm Intel introduced "strained silicon", shocking the rest of the industry. This move improved speed and lowered leakage on transistors in 90nm integrated circuits. An enhanced strained silicon continued the improvement to the 65nm node.
Intel introduced their HKMG (High K Metal Gate) process at 45nm, leaving all competition scratching their heads once again. HKMG made the next node, 32nm possible
In 2011 Intel again shocked the semiconductor industry with the announcement of their TriGate transistor technology. Intel has been shipping TriGate products for two years and has recently moved into production with a 14nm TriGate process. The rest of the industry is struggling with 20nm planar processes with no sign of an answer to TriGate for another couple of years, if ever.
So, Intel has been constantly innovating in process and materials technology for the past decade and this innovation continues to this day.
What is next?
The Holy Grail for the ultimate integrated computing device is on-chip integrated system memory. Length and uniformity of signal paths has been a big issue since the days of the Cray 1 super computer The Cray 1 was a circular configuration in order to make the signal paths as short and uniform as possible.
45 years of semiconductor progress later, Intel is breathing down the neck of achieving the ultimate short signal paths by having all logic and memory on-chip. The company now builds several high performance products with main memory in the CPU package. The next step will be to move the memory on-chip.
This will be no small feat. At 20nm, 4GB of DRAM takes about 320 square mm of silicon, much too large to incorporate on a CPU chip. Two more one-third linear shrinks would take the size of that DRAM down to about 80 Square mm. Removal of the large pin drive circuits would probably further reduce the DRAM size to 60 square mm, well within the ability to integrate on-chip with a very powerful X86 processor.
The problem with this is that the consensus of the semiconductor industry is that there is not two more shrinks left in DRAM. Intel has proved that the death of Moore's Law has been premature for the past seven node shrinks. If anyone can pull a rabbit out of the hat on 10nm DRAM integrated on to a CPU chip it will be Intel.
What happens at this point? Obviously performance, at the same clock rate, goes off the charts. So this architecture will serve the high performance computing applications like servers and super computers very well. On the other hand, the higher performance will allow a lower clock rate with the attendant huge decline in power consumption for the mobile users. Basically everyone from low power users to hyper performance users gets what they want.
Applied to mobile, this system memory on-chip approach can occur even earlier, since the amount of memory is about one fourth the amount used in a PC. For example, an Apple (NASDAQ:AAPL) iPhone only uses 1GB of DRAM. Given the above magic, that amount of DRAM might only occupy 15 square mm of silicon and be easily integrated onto an X86 application processor.
Up to now, mobile application processors have been SoC (System on a Chip) using analog and digital on the same chip. So, we have a mixed mode AP and separate memory to make up a mobile system.
What if the SoC were to be totally digital, including CPU, memory, and the logic parts of communications control and the analog and RF functions were combined on a second chip? An all-digital chip is easier and cheaper to make and to continue shrinking and the analog chip can be made on older, larger node, cheaper, planar processes. In the end, the mobile system is still in two chips that are each optimized for performance and low power consumption.
In summary, Intel continues to innovate, the PC is anything but dead, mobile will belong to Intel when memory moves on-chip, and Intel will prove to be the "Elephant That Can Waltz."
We always used to say that the easiest sell is to existing customers or segments. In Intel's case that would be PCs…in whatever physical format that might take. Today, 300 million traditional PC CPU chips with memory on-chip would produce over $50 billion in high margin sales. Servers and HPC chips another $15 billion and a billion memory-on-chip mobile parts another $30 billion.
And all the capacity to make this stuff is already in place and paid for.
Disclosure: The author is long INTC. The author wrote this article themselves, and it expresses their own opinions. The author is not receiving compensation for it (other than from Seeking Alpha). The author has no business relationship with any company whose stock is mentioned in this article.