Stock Market, Stock, Invest, Trade, Investing, Trading, Financial, Analysis, Articles, Opinions, Discussion, Derivatives, Calls, Puts, Risk, Growth, Leverage
GrassRootsAnalysts.com: Investment Opinions from Independent Thinkers
Contributor Article: Technology / Semiconductor / Memory, Storage, Processors / Intel, Micron Technology
Through Silicon Vias Might End The Reign Of The SoC, Replacing It With The "Ball". But If So, When? Part 1.
January 2014. Minor refinements through 29 May 2014.
Micron (MU) and Intel (INTC) might foresee the end of the era of the System on a Chip (SoC). They might perceive that Through Silicon Vias (TSV) will enable them to fabricate, with superb yields and economy, the "Ball", which should dramatically outperform the finest SoCs at substantially lower fabrication prices. This would be a truly remarkable advancement, and it might actually happen. But it all depends upon when TSV's become genuinely ready for prime time, and that key factor seems shrouded in mystery.
However, my sense is that a process for highly efficient fabrication of tiny reliable TSVs is inevitable - it's just a question of when. It might be rather distant, but it's not impossible that it's imminent or has already been achieved. Special insight into this might give an investor terrific entry timing. But lacking that, it might be wise to invest in Micron and Intel now.
The Ball, like the Hybrid Memory Cube (HMC), represents a leap from two dimensional to true three dimensional silicon structures. And my guess is that Intel and Micron will be first to mass produce it, will dominate the technology for a significant period, and that Micron's role in the venture will propel another round of corporate success which will prove to be comparable in significance to their Elpida acquisition. I can't suggest when this will happen, but my sense is that it's inevitable.
A little history and some technical details will help us understand what the Ball is and why its introduction will prove to be such a watershed event:
Long ago computer scientists and physicists envisioned the future of maximum performance processing in the form of a "Smoking Hairy Golf Ball" (SHGB). It would be a liquid helium cooled superconducting three dimensional solid mass of extremely tightly packed microscopic sized transistors interconnected as intimately as possible into Mother Nature's most space efficient container, the sphere. It would smoke due to vaporization of liquid helium coolant, and would be hairy due to numerous wire connections to its surface. And it's processing speed would be phenomenal mostly because its ultra tiny transistors would be as tightly packed into a perfect sphere as possible, thus minimizing component to component connection distances.
It was a brilliant idealization which, other than the hair element, remains conceptually sound to this day.
The SHGB concept was born of an understanding of this immutable problem: The Cosmic Speed Limit, the speed of light in a vacuum, is 299,792,458 meters per second. Period - Mother Nature slammed her gavel down on this matter about 13.7 billion years ago, and will hear no appeals. The practical speed limit for electromagnetic propagation in solid or mixed materials is very roughly 170,000,000 meters per second. That may seem fast, but in the world of semiconductors it's sluggish - electronic signals can travel only about the length of your index finger in the time a 2 GHz processor completes one clock cycle. Functional blocks are occasionally forced to wait for signals from other functional blocks, especially when those other blocks are located in separate chips (even if in a multi-chip package), thus wasting precious time. It's a very serious problem - it constrains our ambition to develop ever faster and more potent electronic devices.
Natural Laws can't be violated. But when they impose a frustrating constraint we try to do other things to minimize the effect of the constraint. In electronics, the way to minimize delays imposed by the Cosmic Speed Limit is to make components as tiny as possible and place them as close to each other as possible. I refer to this as Intimacy Technology (ITech). The more intimate components are, the faster they can communicate. And at lower power levels too - more distant communication imposes not only noxious time delays, but also higher power requirements because parasitic elements, which are of higher magnitude with longer distances, must be driven with sufficient power to overcome the signal drain they cause.
ITech's delivered a lot already - squeezing more and more components and functional blocks into SoCs has given us very impressive performance. Ever tinier fabrication process nodes are a key element of that advancement of course. But the SoC solution's constrained by serious limits involving fabrication yield, differing fabrication processes for differing circuit classes, and, ultimately, delays caused by the longer distances imposed by a two dimensional environment - I suspect that even within SoCs some circuits must occasionally wait patiently for certain signals which must cross a significant portion of the expanse of the chip to reach their destination. Modern devices couldn't clock in the GHz range without advanced ITech, but this technology can't reach its zenith in two dimensional structures. The SoC's a remarkable human achievement, but it's no SHGB. Nor is it a System Cube (SC) or a Ball. (I'll explain what those are shortly.) The SoC's entrapped in a two dimensional world, and thus forever performance limited.
ITech's Holy Grail is to pack all silicon components as tightly as possible into Mother Nature's most space efficient container, the sphere. That's the genesis of the SHGB concept - that's why the SHGB is round. And that's why we might be headed for a modern version of it, the Ball.
The Ball is profound - it's a dramatic practical advancement which most efficiently minimizes the performance constraints of Mother Nature's immutable speed of light barrier and thus sets a performance standard no provincial two dimensional technology can match. It's ITech taken much closer to its limits. It's the classic SHGB, but evolved to a practical modern form: The Ball won't require liquid helium cooling (it won't be a superconducting device), and thus won't smoke. And it won't have hairy full perimeter wire connections, but rather just tiny bump connections on one or two sides, or on one or two extension layers, or roughly similar. And it'll be much smaller - perhaps modest marble sized for extreme performance systems, and smaller for more ordinary systems.
And it will dominate - the Ball will simply outperform all other systems in speed, overall capability, energy efficiency, and economy of fabrication by substantial margins. Every end product OEM will crave it and be forced by competitive pressures to utilize it. Volumes might be very high, and margins for the first fabricators very lucrative. Because everyone will need and want the Ball.
In my estimation Micron and Intel are now racing toward development and commercialization of the SC and Ball (aka System Ball (SB), but I suspect most will simply call it the Ball).
Micron's Hybrid Memory Cube (HMC), the first commercial true three dimensional solid silicon structure as far as I know, is the harbinger of the Ball - in my view the existence of the HMC strongly suggests the SC and Ball are coming. (I consider this evolutionary connection in more detail in part 2 of this pair of articles.)
The SC and Ball will be true three dimensional structures like the SHGB and HMC, and thus should fundamentally and significantly outperform provincial two dimensional System on a Chip (SoC) technology. To understand why, consider their structure, which honors the conceptual foundation of the SHGB:
The HMC is composed of a stack of naked memory chips plus a controller chip all placed directly on top of one another like a stack of pancakes, and interconnected to one another by means of Through Silicon Vias (TSV). Generally SCs and Balls will be stacks of naked processor, memory, storage, GPU, radio, and ancillary chips - full systems fabricated into a solid three dimensional final form. They follow by logical extension from the HMC, using the same fabrication methods to achieve extremely tight three dimensional silicon structures.
We'll see SCs first - they'll be cuboid implementations which will precede the Ball. Then, to wring out even more performance in the quest for perfection, many SCs will evolve into optimized Balls - each chip will be sized so as to create a final stack which is generally spherical in overall geometry (with step function gradients). It will be an object of supreme technical beauty and stunning performance, yet be very efficient and economical to fabricate.
In the SC and Ball, each chip in the stack will be size and complexity optimized so as to be very easy and cheap to fabricate at superb yields. Each will connect with other chips in the stack over extremely short distances through TSVs. In time the stack will incorporate nearly every semiconductor related function which is central to a product's electronic needs. And the first SCs and subsequent Balls will be dramatic products which drive successful corporate fortunes.
Micron states "With HMC, you can move data up to 15 times faster than with a DDR3 module and use up to 70% less energy and 90% less space than with existing memory technologies." In my estimation the Ball won't initially eclipse the performance of the finest SoC based systems by such large margins. But I do think the margins will be quite substantial. And that the Ball will ultimately be cheaper to fabricate. You can imagine the industry implications if Intel suddenly introduced an SC or Ball with even just 30% speed, energy conservation, and physical size advantages over the best SoC based systems, and could mass fabricate it with superb margins.
But neither the SC nor Ball can exist sans TSVs. And the Micron / Intel joint venture seems to be the only significant entity which can fabricate TSVs in a sample production environment at this time. In my technical opinion the semiconductor world, whether largely conscious of it or not, is accelerating toward the Ball. By virtue of their TSV acumen, the Micron / Intel joint venture seems to substantially lead this technology sprint - in my estimation they're consciously racing toward mastery of production of the Ball. And so far as I know they're the only substantial firms which are pursuing this technology seriously.
TSV technology isn't mainstream yet, but Micron is fabricating it - TSV is the key technology behind the HMC. But frankly I have precious little idea how refined the current process is - fabrication costs and chip space consumption aren't clear to me. However, there are no Natural Law barriers standing in the way of efficient fabrication of tiny TSVs, and the competitive pressure to succeed is enormous. So my sense is that highly efficient TSV fabrication is inevitable. And perhaps Micron and Intel alone fully understand the profound importance of TSV and are working very hard to improve its fabrication until it becomes a highly efficient, cheap, and real estate stingy process - a very high yield and superbly optimized but proprietary process. And it's not impossible that they've already achieved this.
However, it's also possible that it's still terribly difficult, and thus TSVs aren't yet ready for prime time, and that progress in breaking down technical barriers has partially stalled for the time being. The Micron web site states that HMC samples are available now. But I know of no production scheduling.
Bear in mind that the HMC can't simply drop into conventional systems to replace provincial memory - systems must be specifically designed to accept the HMC's unique physical form. And, in spite of the quite dramatic performance advantage the HMC provides, that's not likely to happen until product manufacturers are convinced the HMC's truly scheduled for mass production. And I don't see any evidence of that yet. So maybe a tough technical difficulty is preventing TSV from making a transition from sample production to full production.
However, since achieving a capability to fabricate tiny TSVs at high yield and low cost has profound implications for the doom of the SoC at the hands of the SC and Ball, such news might be managed as a high corporate secret. Perhaps it's even viewed as important enough to artificially stall full introduction of the HMC until Intel and Micron are ready to surprise their competitors by abruptly introducing the SC and Ball.
Samsung's interesting to consider in this affair. They aren't stupid - they're technically astute and might perceive the same future. And there's enormous fire in Samsung's belly - they have no intention of being left in the dust if they can possibly avoid it. So they might be toiling day and night behind tightly closed doors to perfect their own TSV and Ball fabrication processes. That's certainly possible. But I know of no other credible entries in this race. And I'm only speculating about the possibility that Samsung's a serious contender.
Micron's HMC is effectively a partial SC. Since it's a TSV interconnected three dimensional cuboid, it offers dramatically faster system memory performance at dramatically lower power levels and with much smaller physical space consumption. It's a stunning technical achievement which will have a considerable direct impact in high performance IT markets. But the profound implication of the HMC is that the SC and Ball will follow, and will share some measure of the HMC's dramatic performance advantages, displacing today's SoC.
The SC and Ball are the elephants standing directly behind the HMC. It strikes me as odd that the investment community doesn't seem aroused by these elephants in the room yet - the HMC heralds the arrival of the first SCs, followed by the Ball - at some unknown date. And, in my estimation, superb corporate success stories will then follow. Lacking more definitive timing information, in my view now is the time to invest in these opportunities.
The SC and Ball might prove to be another dramatic success story for Micron - perhaps, leveraged from their new very strong fabrication infrastructure and talent position in the industry, and in partnership with Intel, a truly stunning success story.
Or so it seems to me. Of course, prognostication is a perilous endeavor, and very nearly always inaccurate in some measure. There will be lots of intrigue and drama along the way. Samsung might be a strong contender in this race for example. And of course unforeseen events will almost certainly unfold.
But there are two important foundations we can absolutely count on:
None of Mother Nature's Laws will be broken, period, irrespective of need or effort. But when Natural Law barriers don't stand in the way, and intense competitive pressures exist, technical barriers will be overcome in time. A superbly efficient method to fabricate tiny TSVs isn't prohibited by Natural Law barriers, and they are key to an immense industrial treasure. So a method will be devised. Or already has been.
And industry and consumers continue to lust for ever higher performance from their electronic devices. As impressive as the latest products are, we still want faster response, more storage capacity, more advanced displays, longer battery life, easier and more precise information entry and system control, broader application capability and versatility, improved reliability and ruggedness, and higher performance in every other regard - in every realm, insatiable consumers always want more and better. (And increasingly complex software often demands more.) And given similar priced competing products, we select the higher performance option. Such is human nature. The SC and Ball won't change that. But they will move us significantly toward the impossible ideal of the totally fulfilling device. So we'll lust for them - we'll all want an SC or Ball in all our devices. And the first firms to mass produce them will enjoy striking success.
All just in my personal estimation. But refer to part 2 of this article set if you're interested in more technical detail about why I believe the HMC will invariably lead to the SC and Ball.
I'm thrilled to be intensely invested in Micron common and calls. I was especially aggressive in accumulating 17 January 2016 $27 Micron calls in earlier times, and they've already been wonderfully lucrative. I hold similar Intel calls too, and continue to add to those positions. And my guess is that their growth is effectively just beginning - I suspect I'm invested heavily and with considerable leverage in a dramatic success story which seems to be progressing well and will ultimately eclipse the Elpida acquisition story, yet is popularly viewed as if in nascent development stages, and whose profound importance seems generally unrecognized by the investment community. For Micron and Intel, I suspect the best is yet to come.
If I have that about right, this is a superb time to invest in Micron or Intel. A seasoned investor's instinct is to bail from a firm when great success stories, such as the Elpida acquisition and very healthy consolidation of the memory and storage industry, become mature and universally recognized - instinct yells from the mountain tops that the peak of the reward curve is at hand, and thus it's time to sell. But this time lightning might strike twice for Micron, and with an even more powerful second bolt, much as it did for Apple (AAPL) after their success with the iPod. This time it's different. Again...
These are simply my personal observations, opinions, and projections. We all try to extrapolate to divine the future, but we can't actually see the future of course. All investors must decide for themselves how to manage their precious money - everyone must steer their own ship.
My heartfelt sincere thanks to the superb array of independent Micron and Intel analysts. Though a very long term Micron investor, I intensified my concentration in Micron common and calls enormously in response to the impressive array of world class smart, experienced, energetic, civil, and good humored Micron and Intel commentators. I've benefited greatly, and am in their spiritual debt. And I hope they'll continue to focus their remarkable skills on the continuing story, especially if I'm somewhere in the very roughly correct ballpark in judging the approaching SC and Ball drama. If so, we'll all have the thoroughly analyzed and constructively debated information we need to steer our Micron and Intel investments to, good ancillary luck allowing, truly embarrassing levels of success. That wouldn't provide immortality, alas, nor even much change the most fundamental joys in life. But it could certainly sweeten some aspects of our one and only very brief ride on the consciousness trolley.
Best of luck to all, Bruce Campbell
Relevant disclosures: Intensely long Micron common stock and call options. Significantly long Intel calls. Usually modestly long Intel common. Occasionally modestly long ChipMOS common or call options. All are long positions, and none are hedged other than quite modest precious metal common holdings. But I might divest all but my PM holdings if a serious systemic market collapse seems imminent.
This original composition expresses my personal investment views. I have no business relationship with any entity in which I hold investments. On about 19 January 2014 an earlier second party modified version of part one of this article set appeared very briefly on another web site, but I quickly retracted it due to strong displeasure with the modifications and their means of implementation.
Copyright 2014 H. Bruce Campbell, Creative Commons license: Attribution (BY) + Noncommercial (NC) + No Derivatives (ND)
NoSpam Notice: Unsolicited Commercial Email (UCE), "Spam", or any unsolicited subscription based email distributed on an "opt out" basis is absolutely prohibited. Do not ever send any such email to any GrassRootsAnalysts.com address.
Report mail misconduct to UCE@FTC.gov.