Innovation Monitor: Gartner Hype Cycle Trend #4 — Beyond Silicon
Welcome to this week’s Innovation Monitor. This week, we’re diving into the arms race around the hardware that powers our devices and machinery. Silicon has been the computing industry’s lifeblood for decades, guided by Moore’s Law, which forecasted the trajectory of advancements in computing power that fits onto a circuit board.
Today, however, engineers are approaching the physical limit of how many transistors they can pack on a chip. Prominent computer scientists like parallel computing pioneer Charles Leiserson have pronounced the end of Moore’s Law.
In February MIT Technology Review noted that even if “chipmakers can squeeze out a few more generations of even more advanced microchips, the days when you could reliably count on faster, cheaper chips every couple of years are clearly over. That doesn’t, however, mean the end of computational progress.”
Others like economist Neil Thompson believe the focus on specialized chips is causing “chips for more general computing [to become] a backwater,” slowing the overall pace of computing improvement. In fact, we might soon be citing Huang’s Law, WSJ tech columnist Christopher Mims’s term for “how the silicon chips that power artificial intelligence more than double in performance every two years.” Huang’s Law implies both hardware and software innovation.
But there is a world beyond silicon. Which that brings us to Gartner’s third hype cycle trend.
And this is where things get a bit weird, in a wonderful kind of way. You may be familiar with the concept of quantum computing by now (and if you’d like a lucid deep-dive, here’s an excellent guide, along with one of our previous newsletter editions!). So let’s focus on two other key emerging technologies.
DNA computing and storage has resurfaced in academia thanks to advances in synthetic biology — advances quite analogous to the computing industry, in fact. Infrastructure like low-cost genetic sequencing, automated cloud-accessed laboratories, and biology-as-a-service may evolve into something that democratizes the biological “app” development process, the way Apple and Android’s app stores and SDKs have done for software.
For the final section, we’ll review carbon nanotubes. Though they may not seem less riveting than qubits or DNA hard drives, they have a host of transformational properties that, in theory, could spell faster, more efficient chips. But as MIT PhD student Mindy Bishop notes, “it’s one thing to know those benefits could exist one day, and then there’s all the work to make it happen.”
As always, we wish you and your community safety, calm and solidarity as we support each other through this unprecedented time. Thank you for reading!
Erica Matsumoto DNA Storage & Computing
“From a resiliency and storage density perspective, nothing beats DNA. Properly stored, DNA can last for at least 500 years. And a gram of DNA can store over 200PB of data,” writes Gartner. Researcher Michael Chui said that “one kilogram of raw DNA could store all the world’s data today,” according to Fast Company.
Though the idea sounds complex, the concept behind DNA storage and computing is not. Digital content is mapped to four nucleotides (adenine, thymine, guanine, and cytosine). Each nucleotide represents two bits. These nucleotide codes are then used to create synthetic DNA, which is replicated and stored in DNA strands. These strands are copied millions of times to make reading data easier.
Gartner notes “despite two successful prototypes, [DNA storage and computing] is currently rudimentary and expensive with significant technical barriers to mainstream use. However, the impact of a successful DNA computing and storage option would transform data storage, processing parallelism and computing efficiency.” For one, it would be much, much cheaper to store data. Take the Large Hadron Collider, which produces petabytes of data every year. This is stored on magnetic tapes that need to be changed every decade or so. A DNA-based dataset could be replicated millions of times and at a lower cost.
What about processing? That’s where enzymes come in — “proteins that act as catalysts to perform a logical operation on a collection of DNA.” Custom-designed enzymes could act like logic gates that read data and output new DNA. But most experiments with molecular computing haven’t yielded flexible results. “Using DNA to compute is ‘like having to build a new computer out of new hardware just to run a new piece of software,’ noted computer scientist David Doty last year.
Doty and his team set out to build something much more reprogrammable. Wired wrote that “Doty and his colleagues from Caltech and Maynooth University demonstrated just that. They showed it’s possible to use a simple trigger to coax the same basic set of DNA molecules into implementing numerous different algorithms. Although this research is still exploratory, reprogrammable molecular algorithms could be used in the future to program DNA robots, which have already successfully delivered drugs to cancerous cells.” The Wired piece also has an excellent explainer of the DNA computing process, if you want a deeper look. Graphene & Carbon Nanotubes We typically confine the notion of “2D” to flat images — paper drawings, 2D video games, paintings. But scientists actually consider some materials to be two-dimensional: single sheets of atoms, or materials a few nanometers thick — graphene is a popular example (check out this WEF post for a simple explainer on the material). A carbon nanotube (CNT) is basically a 2D sheet of graphene wrapped into a cylinder — though there are plenty of structural variations.
One reason researchers are looking at silicon alternatives is that as things shrink down to the atomic scale, they get… unstable. Electrical current starts to leak and the heat released degrades efficiency. CNT transistors, according to researchers, are faster and up to an order of magnitude more efficient. But CNTs come with their own host of problems:
“One problem is that when carbon nanotubes are made, they come in two types mixed together: the first are semiconductors that are perfect for creating integrated circuits, but the second conducts electrical current like a wire, which sucks more power and can even undermine a circuit’s performance…. Another problem is that to make the chips, a uniform monolayer of carbon nanotubes needs to be deposited over a wafer. But this has proven hard to do because nanotubes have an annoying tendency to bunch together. A bundle of them that lands on a transistor can knock it out of action.”
Last year, a group of MIT researchers created a 16-bit microprocessor with over 14k CNT transistors. The chip ran a “Hello, World” program successfully. But CNT-based chips would need to be to scale to billions of transistors to run complex software.
This month, a team from Harvard and the Samsung Advanced Institute of Technology reported that they had managed to fit 1,000 2D-material-based transistors on a chip. This enabled the researchers to run a functional neural network on the chip, which could recognize over 1,000 handwritten digits with 94% accuracy. “This is the first demonstration of a neural network with two-dimensional materials that can interact with light. Because it computes in memory, you don’t need separate memory and the calculation can be done with very low energy,” said Houk Jang, one of the accompanying paper’s authors. This Week in Business History September 26th, 1914: The Federal Trade Commission (FTC) is established (from the FTC website):
The Federal Trade Commission was created on September 26, 1914, when President Woodrow Wilson signed the Federal Trade Commission Act into law. The FTC opened its doors on March 16, 1915. The FTC’s mission is to protect consumers and promote competition. As the FTC celebrates its 100th anniversary, our thoughts turn to its unique mission, significant events in Commission history, and its staff, stakeholders and constituents — present and past. On January 12, 2015, President Barack Obama visited the Commission, the first presidential visit to the Commission since 1937.
This email was sent to <<Email Address>>
why did I get this? unsubscribe from this list update subscription preferences
NYC Media Lab · 370 Jay Street, 3rd floor · Brooklyn, New York 11201 · USA