- Show All
Packaging: What's holding us back?
Packaging: What's holding us back?
In June of 2017, just over 170 scientists, researchers, government representatives and high-tech industry experts from 17 countries converged on the Dutch city of s’Hertogenbosch. They came to discuss next generation technologies the world is going to need in 2030 and beyond in the World Technology Mapping Forum. That’s important now because we’re reaching the economic end of what’s popularly known as “Moore’s Law”. And particles of light (photons) rather than electrons will be the engine driving many new applications in communications and life sciences that we’ll soon take for granted.
During the discussions, it became clear that that huge volumes of integrated photonics chips are going to needed in the next few years. Dr Michael Lebby, who conducted a market survey for PhotonDelta remarked, "We noticed that optical transceivers are going to use a lot of more integrated photonics than many have predicted. But we can see some gaps emerging between what the datacenter industry is demanding by 2020 and the price-points that researchers say they can deliver. A lot needs to happen in the areas of yield and packaging”
We’re starting a series profiling the companies that tackling the difficult packaging issues.
Twenty years ago, a typical challenge might have been aligning a single laser diode to a single fiber pigtail. Today we are seeing chips with multiple inputs and outputs, all of which need to be aligned in multiple degrees of freedom. So that nano-alignment process emerges as being absolutely critical. Scott Jordan
PI (Physik Instrumente) & Packaging
During the WTMF, it became clear for the photonics industry to scale faster has less to do with the chips themselves and far more to do with the way the chips are connected up to the outside world. That’s the challenge of “packaging”.
Scott Jordan is Senior Director of NanoAutomation for PI (Physik Instrumente), a private company with headquarters in Karlsruhe Germany and around 1000 employees worldwide. PI has been one of the leading players in the global market for precision positioning technology for many years. They recently completed construction of a €13 million technology centre. Jonathan Marks (Editor-at-Large PhotonDelta) spoke with Scott in advance of seminars they are holding in Eindhoven September 14 and 15th 2017.
Entering a Perfect Storm
Right now, the market is being driven by data centers”, explains Scott “which in turn are all-of-a-sudden being propelled by Big Data, IoT, streaming media, social networks, personalized medicine, autonomous vehicles and so on.”
“If you’re getting the impression that there’s a Perfect Storm of demand that just switched on, then you are getting the picture. Bandwidth consumption has grown monotonically at about a 400% annual rate for the past two decades and this is just cascading and mushrooming now.
Meanwhile from a physics standpoint there are parallel challenges in accommodating the necessary throughput, speeds and sheer capacity together with the energy consumption this all requires.
Depending on the statistics you read, up to 8% of the United States’ electricity is consumed by data centers, far more than lighting by now, and lighting isn’t growing exponentially. This is clearly not a sustainable trend!
Meanwhile, the physics of computation show that most of the energy consumed in computing is actually in the conveyance of bits from place to place. Charging up a long wire of copper to convey a bit from point A to point B takes energy, and lots of it by now. It is significantly more energy efficient to use photons rather than electrons for such purposes.
All of this is on top of the benefits that multiple parallel data streams, speeds, capacity, etc. that optical communications can bring. The striking thing about photonics is the continuous diminishment of scale. In the “big photonics” boom of twenty years ago, it was first all about replacing satellite links with faster/lower-latency/higher-capacity transoceanic and transcontinental links. Then came regional, and metro, and fiber-to-the-home, and so on. Now it’s all about connectivity between data centers, between racks, between boxes and between cards in the boxes. The next step is to improve data rates between chips, between cores…
The trend is clear: Scales keep compressing.
That trend has been monotonic and unrelenting. And there’s another trend: as the scales diminish, the quantities rise in some inverse power-law. So suddenly the number of devices that need to be manufactured absolutely skyrockets.
That is reason why Silicon Photonics is happening, and happening now. There’s no other way to scale the quantities, just as there was no other way to scale manufacturing at the dawn of the digital age besides making microprocessors out of silicon. The former ways of doing things (with printed circuit boards, wire-wrapping, solder etc) simply would simply not have done the job.
And remember, we’re only talking about connectivity! There are other areas that are growing like quantum computing, quantum communications, optical logic and so much else.”
Jonathan: "80% of the costs are in packaging not in the cost of the chip" is often heard. What has PI developed to tackle this problem and why are you labelling it as unique?
That 80% number is typical of microelectronics. I don’t think anyone really wants to admit how bad that ratio is for SiP right now! But that is the problem we are solving.
Getting photons out of a sender (a laser, a chip…) and into a receiver (another chip, a fiber…) takes nanoscale-precise alignment of the devices. Twenty years ago, a typical challenge might have been aligning a single laser diode to a single fiber pigtail. Today we are seeing chips with multiple inputs and outputs, all of which need to be aligned in multiple degrees of freedom. Those inputs and outputs can interact, too— so alignment can be a moving target for that reason, plus there are physical drift processes from thermal changes, and so on. A difficult, multivariate, unpredictable situation!
And it starts at the wafer. Since packaging is so costly, it doesn’t make economic sense to package bad devices. So, the devices must be tested while they’re still on the wafer, thus ensuring that only good devices proceed through that costly packaging process. And then at every node in the packaging chain, the health of the devices needs to be re-certified.
So that nano-alignment process emerges as being absolutely critical, since it must be performed multiple times from the wafer forward: for testing, for multiple packaging steps, for multiple test steps during those packaging steps…
Only a handful of key players
The photonic-alignment field has been a small one, with perhaps five or 6 key players throughout its history. It’s a small club which PI is a part of. Early alignment automation allowed single couplings to be optimized. In multichannel and multi-degree-of-freedom situations, this meant that you’d do one alignment, then a second one (perhaps on another channel, or on an input and then an output, or in a different degree of freedom). The problem is the second alignment would screw up your first alignment so you’d have to back and repeat the first one. It meant constant tweaking this and tweaking that. Ultimately, after some considerable period of valuable production time, you’d have all the necessary bits and DOFs aligned. But then, if something drifts you’d have to do it all over again.
Clearly this is not sustainable!
What’s unique about our FMPA technology is that we’re able to do all of the above in parallel. So, it’s one fast step rather than a lengthy loop of sequential steps.
This is totally unique and new, and frankly many people wonder how it can possibly be since there are geometrical and optical interactions between the channels, inputs/outputs and DOFs. Suffice it to say we have cracked the math, so that all those alignments can be done in one go. The resulting productivity improvement can exceed two orders of magnitude. That’s not hype. That’s the physics of it.
For SiP vendors, it’s one of those things that must be adopted, since your competitors will adopt it and bankrupt you if you don’t.
Jonathan: There seems to be a trend whereby application makers need to think about the packaging aspects at the same time they are designing the chip. And is there not a problem of a lack of industry standards?
Yes and yes. The complexities (and practicalities) of packaging caught a lot of these players by surprise. The necessity for testing at the wafer-level did too!
These are still early days, with a lot of innovation and churn happening. “Lack of industry standards” is understandable since the inventions are still coming rapidly. Industry standards are for mature industries, not for jackrabbits. Sure, they’ll happen, and they’ll make life easier, once the blistering pace of innovation cools off. I don’t see that happening anytime soon.
Jonathan: Does your process involve real-time testing of specs?
It enables that, among other things.
Jonathan: We see some companies having to test all their incoming chips because of challenges involving yield. How does your system know it has done an accurate job?
We know alignment is complete when the gradient-search process drives the alignment error (= the gradient) to zero.
There are more technical explanations in an article I wrote for Photonics Spectra for https://www.photonics.com/Article.aspx?AID=62191
Gradient search technology has been around for a long time (I invented the digital gradient search in 1987); what’s new here is the ability to perform an arbitrary number of gradient searches on a device all at once. But the physics of it is the same as it was in 1987: you deduce the local gradient of a coupling and drive that to zero by aligning.
Jonathan: But PI’s heritage is in silicon. But do these techniques translate well for other materials? What do you need to adapt? What are the challenges in building nano-robots?
Our nano-robotic, fully parallelized alignment technology that is proving so essential to the manufacturing of Silicon Photonics is by no means limited to "Silicon" photonics devices and technologies. The same exacting alignment challenges await with other materials and approaches. In fact, already we have several customers in the quantum computing field, including some devices that are astonishingly close to actual manufacturing in volume.
Jonathan: Can you give me some background on the company and how you decided to specialize in this part of the photonics industry?
Personally, I fell headlong into alignment automation thirty years ago, when the trick to doing a practical digital gradient search emerged. Things moved quickly from there. PI was involved in the market from the beginning, back when “photonics” was a separate market entirely unrelated to semiconductors. At the same time, semiconductors has historically been PI’s largest market, so when photonics and semiconductors started to merge, we were well positioned to draw from both fields to help our customers innovate.
That bring me to a point I try to make often. Innovation occurs at intersections, and since PI plays across a multitude of fields (semiconductors, photonics, life sciences, advanced microscopy, genomics, defense, materials sciences, data storage…), this puts us in a unique position to help our customers cross-pollinate across fields.
It is glorious to witness this unfolding in real-time, and that’s what you see in Silicon Photonics today.
Jonathan: But can you name examples (or sectors) where your technology is already working and to what extent is it ready (or working) in volume production? There seems to be a vast gap between volume production in electronics and photonics.
Everything I’ve mentioned: wafer probing, chip packaging and test, quantum development (and, soon, manufacturing), and so on.
It’s really happening. All of it.