This BLOG is only for the geeks. Please don’t bother to read if you have no idea of Nanometers.
A few days ago at the IEDM, Intel disclosed their future process node technologies, showing on what will probably become an iconic slide how Intel expects to have a 2-year cadence for the development of its manufacturing process node technology and that it also aims to reach the 1.4nm node by 2029. Naturally, as usual, Intel described how challenging it will be to get to such an advanced node and how hard the company is working to reach it.
It is very interesting to note that according to IRDS forecasts in the year 2033 we should reach the 0.7nm node based on Vertical Gate-All-Around (VGAA) FETs with 6nm diameter and 14nm pitch, which will probably be the final limit of the Moore’s Law. When we compare Intel’s proposed goal for 2029 with IRDS forecasts for the continuation of Moore’s Law, it’s possible to see how reckless Intel’s evolution proposal is, especially when compared to the much more modest and short-term roadmaps proposed by competing foundries like Samsung and TSMC.
The way Intel made this disclosure reminded me of a situation that happened 20 years ago when even though there was a clear warning that Intel’s plans might not materialize as expected, Intel still insisted on making a bold prediction for the future of its technology and eventually had to face the harsh reality of the laws of physics.
In 1999 an Intel researcher published an article in Science Magazine showing fundamental physical limits to the continuation of Moore’s Law, with no known solutions, and that these would represent a real challenge to the semiconductor industry, one never before faced. This was something unheard of, up to that point no Intel researcher had ever made any such claims that Moore’s Law could fail.
Even though this was a clear sign that the challenges ahead could prevent Intel from reaching its goals, at the end of 2000 Intel predicted they would be reaching 10GHz by 2005 using less than 1V, a very bold proposal for the time.
However when the industry finally entered the sub-100nm regime, the death of Dennard Scaling caused a complete paradigm shift in the semiconductor industry. Later with the arrival of the Frequency Wall, which basically limits the operating frequency of processors to 5GHz, along with the emergence of the Dark Silicon phenomenon (I will have a specific article covering this in the future) contributed to the complete obliteration of Intel’s golden dreams. Intel never came close to achieving the company’s goals. Never has an Intel processor been able to get close to 10GHz without the use of liquid nitrogen for processor cooling and a nuclear power plant.
Over the past few years clear signs have emerged pointing to Moore’s Law having its days numbered. Many of the challenges in continuing the historical trajectory of performance improvements with the continued evolution of CMOS are proving extremely difficult or too expensive to overcome. With R&D spending rising since the 2000s at a considerably higher level than in previous decades, the number of companies with integrated manufacturing facilities developing bleeding edge manufacturing technologies shrank from 29 in 2001 to just 4 at the beginning of 2018, and finally to just 3 in the present day with GLOBALFOUNDRIES stopping the development of its 7LP node. With these and many other clear warning signs, the semiconductor industry as a whole has focused its research efforts on finding alternatives.
Since the late 2000s, the semiconductor industry has been developing “More-than-Moore” solutions (integration of heterogeneous functionality in the CMOS platform) and “Beyond CMOS” research (ranging from new computing elements to entirely new computing architectures that are at least initially compatible with CMOS) as an effort to prepare for the ultimate end of Moore’s Law, when it is no longer physically or economically possible to continue the scaling process. Intel has also made great strides in developing research within these trends, developing exquisite solutions across these various fields of research.
Among Intel’s many advancements we can highlight EMIB and its variants, which allows Intel to develop 2.1D interconnects in an advanced package at a lower cost than using an interposer. There’s also Foveros, which enables Intel to develop 3D-IC design, and more recently Intel MESO, a new device resulting from Intel’s “Beyond CMOS” research which has enormous potential for the future. Also Intel is developing on-package photonic interconnect (something I discovered through my search for Intel patents), which will allow Intel to develop a photonic NoC (Network-on-chip ) in the near future. Much effort has been expended over the past decade in developing these technological advances to secure a lifeboat when Moore’s Law finally sinks.
However, that did not stop Intel in 2017 once again trying to keep Moore’s Law alive through an unprecedented effort. On its Manufacturing Day, Intel announced that it would maintain its efforts to extend Moore’s Law by introducing the hyperscaling process, placing even more difficult scaling goals, and betting high on its 10nm node, which promised a glorious return of Moore’s Law on track, with similar yields to 14nm and a very promising future.
What has been the result of such efforts? Search on Google for “Intel 10nm meme” or “Intel 10nm fiasco” and you will quickly understand that this story did not have a happy ending. It’s late 2019 and 10nm Intel is no better than its 14nm and it definitely does not present a future as bright as Intel dreamed. Today Intel is having a hard time, seeing its opponents reaching denser nodes and at a much faster evolution cadence. Although foundry nodes are evolving rapidly, it must be made clear that both TSMC and Samsung are also at a crossroads, in a desperate search for alternatives to survive the impending end of Moore’s Law.
Also, at the IEDM, on the same slide in which Intel proposed 1.4nm by 2029, Intel also proposed that along the way there would be the possibility of architectural back-porting from an advanced node to an older node, clearly showing no such confidence as in the past that it can achieve the promised advances for the future.
Looking at this scenario, we must ask: Is it really necessary that Intel keep Moore’s Law alive at all costs to secure its future?
Intel’s history is marked by pioneering and technological innovation. Over the years Intel has struggled to keep Moore’s Law alive, often setting aside alternatives that could allow Intel to be in a more comfortable situation and with a brighter future. As much as Moore’s Law has been in the past a synonym with innovation, it is necessary that Intel recognizes its end. There is a bright future with countless possibilities for semiconductor evolution, and Intel needs to leave the past behind and embrace the future.
Source & Further Reading: