The first time you see it, it doesn’t feel like the future at all. It looks like a relic: a milky panel, no glow, no pixel shimmer, no frantic refresh. It sits there, motionless and pale, like frosted glass that has forgotten how to be transparent. Then, with a soft, almost shy delay—a bloom of shapes begins to appear. Grey strokes swell into view, darkening as if ink is seeping upward against gravity. A number. A line. A pattern. No light is shining out of it. No fan is humming behind it. And yet, something is thinking.
The technology that refused to die
This ghostly panel is not a new invention. It’s the quiet return of an old one—an analog memory technology first explored more than fifty years ago, when computers still filled rooms and “portable” meant you could maybe shove it onto a truck. The idea was simple and radical at once: instead of storing information as flickering electric charges in silicon chips, why not store it physically in materials and patterns that remain in place even when the power goes out?
For a while, the world flirted with such ideas—optical memories that used light, magnetic films, even strange hybrids of chemistry and electronics. They were bulky, finicky, and took a back seat the moment digital semiconductors began their relentless rise: faster, smaller, cheaper, everywhere. Analog memory sank quietly into the footnotes of history.
But history, as we’re discovering, is full of seeds that sprout late.
In laboratories across China, engineers and physicists have dug back into that forgotten toolbox and pulled out an approach that suddenly feels uncannily right for our overheated, overpowered, always-on age. They’ve taken a half-century-old concept—storing information not as streams of electrons but as stable patterns in materials—and married it with modern fabrication and artificial intelligence. The result is an “old-new” computing layer that uses up to 200 times less energy than our conventional digital hardware for certain tasks.
In a world where data centers already emit more carbon than some countries, that number is an invitation: What if we’ve been building the future with the wrong kind of bricks?
When the screen doesn’t shine back
To understand why this revival matters, imagine your daily universe of screens. Your phone pours white-blue light into your face. Your laptop, your watch, the digital timetable at the bus stop—everything glows. Behind that glow is a violent dance of electrons, billions of tiny switches flipping on and off tens of billions of times per second. It is dazzling, and enormously wasteful.
The resurrected technology works differently. Think of it as closer to the way a notebook works than a laptop. When you write something on paper, the page doesn’t need energy to keep those words in place. They’re just there. Energy is spent in the act of writing; not in remembering.
China’s analog revival follows that same philosophy. These devices store data as stable physical states—colors, patterns, charge distributions, or molecular alignments—that don’t evaporate when power is removed. Power is used only when information is changed or read, and even then, far more gently than in the brute-force switching of transistors.
Watching one of these panels update feels strange at first. There is a pause—a breath—before images fully settle into being. It’s like watching a Polaroid photo develop, quietly thickening from nothingness into clarity. We have grown so used to light that pulses and screens that strobe that we have forgotten the sensual appeal of slowness, of things that appear instead of explode into presence.
The strange calm of analog memory
Step into one of the low-lit labs where Chinese researchers are tuning these materials and you notice something rare for a frontier of computing: quiet. There are no server racks with their wasp-swarm buzz of fans and spinning disks. Instead, you see a table, a panel the size of a paperback, a handful of cables, and a monitor showing a graph that crawls rather than jitters.
On the panel, test patterns emerge like frost on glass: lines, spirals, patches of grey that shift in response to faint whispers of voltage. These devices do not need constant power to hold their state. Turn off the supply, come back later, and the patterns are still there, as if memory has become part of the material itself—a tattoo rather than a projection.
Engineers explain that in some experimental setups, a single modest burst of energy can write information that then sits stable for hours, days, or even longer. Compare that with the desperate maintenance needs of modern RAM, which must be fed and refreshed constantly to retain its bits. If RAM is a hummingbird heart, pounding and frantic, this resurrected memory is a tortoise heartbeat: slow, economical, hard to disturb.
This is where the energy savings come from. Many of our most demanding digital tasks—like inference in machine learning, image recognition, or signal processing—don’t actually require the full dance of universal, flexible computing. They can be done by structures that are half storage, half calculation, fixed in place like carved riverbeds that shape flowing water without effort.
How a 50‑year‑old idea suddenly became relevant
Why didn’t this triumph decades ago? Because, back then, the world wanted universality and speed. Early analog and optical memories were impressive but quirky. They could store images and patterns, even perform some calculations in the way their light or charges traveled, but reprogramming them was clumsy. We chose chips that could do anything, at high speed, even if it meant pouring energy down the drain.
What changed is not the physics—it’s the question we’re asking.
In the age of artificial intelligence, we increasingly ask our machines to do a narrower set of things over and over: classify images, detect anomalies, recognize speech, recommend content. These aren’t general-purpose tasks. They’re pattern problems.
Pattern problems are exactly where analog shines. Instead of representing each bit perfectly and then stacking bits into complex models, analog architectures can embed whole relationships and correlations directly in physical form. A pattern of light hitting a surface may be transformed, in a single, nearly passive step, into a classification: cat, car, tree; normal, suspicious, dangerous.
China’s researchers have taken those dusty 1970s ideas about storing patterns in physical media and layered today’s understanding of neural networks on top. The devices no longer simply remember; they compute by remembering. The material itself becomes a network of weighted connections—analog cousins to the learned weights in digital AI models.
This means that once trained, an analog AI chip can run its inference with vanishingly small energy. No massive arrays of GPUs, no acres of air conditioning. Just light, or small electrical signals, passing once through a carefully sculpted medium that has already “decided” how to respond.
200 times less energy: what that really means
Numbers like “200 times less energy” almost sound like marketing slogans. It helps to make them concrete. Consider a tiny AI module in a wildlife camera, strapped to a tree in the Wuling Mountains to monitor elusive golden snub-nosed monkeys. A digital module might sip a bit of power, but over days and weeks, that sip becomes a drain, and batteries must be bigger, replaced more often, or paired with larger solar panels.
An analog module built on this resurrected technology can, in principle, perform the same recognition—Is that a monkey? A human? Just wind in the leaves?—for a fraction of the power. The camera can sleep and wake like an animal, using tiny bursts of energy to process incoming scenes, then returning to stillness. Batteries last longer. Equipment needs fewer visits by rangers hauling replacements through ravines and snow.
Scale that modest savings up: traffic cameras, factory sensors, remote weather stations, underwater monitors, agricultural drones—all doing small shards of thinking without needing full-blown, hungry silicon brains. Suddenly, “200 times less” becomes not a number, but a change in what is possible in the wild world, far from wall sockets and fiber lines.
| Feature | Conventional Digital | Resurrected Analog‑Style Tech |
|---|---|---|
| Energy use for specific AI inference tasks | High, continuous, active switching | Up to ~200× lower; mainly on write/read |
| Data retention | Volatile; constant refresh needed | Semi‑permanent physical states |
| Best suited for | General‑purpose, flexible computing | Pattern recognition, fixed AI models, edge sensing |
| Form of information | Binary bits, clean on/off states | Analog levels, gradients, or physical patterns |
| Environmental footprint | Large data centers, heavy cooling | Smaller systems, lower heat, easier to deploy off‑grid |
From data centers to dragonflies
The biggest shift this “resurrection” hints at isn’t just technological—it’s ecological and cultural. For the last two decades, our mental picture of computing has become increasingly vertical: towers of servers, stacked in frozen halls, devouring electricity. Intelligence, we assume, lives in these mountains of machines, and the rest of the world merely sends them questions.
Analog‑style computing chips whisper a different possibility: flattening the landscape of intelligence, letting it seep outward into the fabric of things. A solar‑powered sensor on the edge of a rice paddy might quietly calculate water stress in the soil by analyzing reflected light, without shipping raw data to a distant cloud. A buoy rocking in the South China Sea might detect illegal trawlers through analog processing of vibration patterns. A tiny unit on the underside of a drone’s wing might make split‑second decisions about flight and wildlife avoidance without the power budget of a gaming laptop.
Seen this way, China’s revival of analog memory is not nostalgia—it’s convergence. Nature has always favored low‑energy, highly distributed intelligence. The dragonfly does not carry a supercomputer in its skull; it carries exquisitely tuned circuits that respond instantly to a handful of critical cues. The mycelial threads of a forest do not run at gigahertz; they run at the speed of moisture and chemistry, and yet they route resources where they are needed, solve problems in their own slow, robust way.
Slow is a kind of speed
There is a quiet irony at the heart of this story. We obsessed over speed: faster processors, higher frame rates, lower latency. And yet the crisis surrounding our technology—its energy appetite, its emissions, its waste heat—comes from that same obsession. Analog resurrection invites us to widen the meaning of speed.
Speed, here, becomes: how quickly can a system achieve what matters, per unit of energy? A massive data center might crank out answers in microseconds, but if most of that work is spent shuttling redundant bits back and forth, its true speed—its speed relative to energy—is poor.
The newly revived Chinese devices, by contrast, move slowly by digital standards. Their updates may take milliseconds, or more. Their resolutions may be modest, their flexibility limited. Yet for the specific jobs they’re built for, they are astonishingly fast in energy terms. They reach “good enough” answers using almost nothing, in a way that echoes natural systems: efficient not in raw pace, but in purpose.
This does not replace digital; it complements it. The future likely belongs to hybrids: digital brains orchestrating, analog senses at the edge interpreting, both working together like cortex and nerve endings. The resurrection of this half‑century‑old idea is less a regime change and more a long‑overdue reunion of estranged siblings.
China’s bet on forgotten tools
Behind the poetic image of ghostly panels blooming with patterns lies a very practical national calculation. China is wrestling with the relentless appetite of AI and cloud computing just as much as any other country. Power grids strain. Coal plants still loom. Data center moratoriums have appeared in some cities, limiting new builds.
In that context, the appeal of a technology that can do useful AI work on a fraction of the energy is enormous. It promises not only smaller electric bills, but strategic resilience: intelligence that doesn’t depend on a handful of hyper‑centralized, power‑guzzling sites.
Chinese universities and research institutes, often working hand‑in‑hand with industry, have begun filing patents and publishing papers on novel analog memory cells, optical neural chips, and “in‑memory computing” arrays that blur the line between where data is stored and where it’s processed. The historical DNA of these projects reaches back to 1970s beam‑steering, phase‑change films, and early holographic memories. But the tools—nanofabrication, machine‑learning training methods, modern materials science—are unmistakably 21st century.
There is also a cultural dimension. In reviving techniques that Western industry largely abandoned, China positions itself not simply as a follower in the standard semiconductor race but as a pioneer along a side trail that suddenly looks very promising. The story of “old tech, new life” becomes a story of technological sovereignty: of finding value in forgotten corners, rather than only where everyone else is digging.
Fragile, experimental—and full of promise
None of this is guaranteed. Early prototypes are fragile. Manufacturing yields are still low. Integrating analog memory systems with the vast digital ecosystem—software tools, developer habits, standard interfaces—is a formidable challenge. And analog’s very nature, its tolerance for gradients and imprecision, can be as much a bug as a feature in a world trained to worship exact digits.
There are questions of longevity: How long do these physical states remain stable? How do they respond to heat, humidity, time? Can you rewrite them thousands, millions of times without fatigue? Can the quirky differences between individual devices be tamed, or perhaps embraced as part of their character, much as biological systems use variability as a resource?
Researchers in China and elsewhere are chipping away at these questions, cycling devices through brutal test regimes, probing their limits with microwaves and lasers, iterating materials like potters trying new clay recipes. The resurrection is real, but its form is still fluid.
A quieter future, humming underneath
Still, it is worth imagining what happens if this bet pays off.
You might not notice it directly. Your phone will still be a rectangle of light. Your laptop will still whirr awake each morning. But somewhere out beyond the edge of your attention, a second layer of intelligence will have spread: small, quiet, analog‑bred minds doing simple work well, for almost nothing.
In a rewilded valley, animal collars ping a base station with short bursts of processed data instead of hungry streams of raw signals. In a fishing village, analog chips listen to the sea’s noise and warn of storms or ships long before human ears can. Along a smoggy ring road, traffic lights adjust to flows calculated not by distant servers, but by tiny analog eyes in each intersection, tuned to the specific rhythms of that neighborhood.
Inside homes, low‑power assistants may finally reach a point of true unobtrusiveness—no constant connection, no giant servers burning away somewhere just to answer a question about the weather. Instead, a modest device on the shelf, humming rarely, its sleep almost unbroken.
And on the back of a dusty shelf in some electronics museum, a yellowed panel—ancestor of this new generation—will sit behind glass. Most visitors will walk past, paying it no mind. A few might stop, read the plaque about a 1970s optical memory experiment, and smile at the thought that the world once tried this path, let it go, and then, humbled by energy and heat and limits, circled back.
Resurrection, after all, is not about going back. It is about returning with new understanding of why something mattered.
That understanding now feels urgent. Our digital civilization is bright and loud and astonishing, but it is also hot and hungry. In resurrecting a 50‑year‑old, analog‑leaning technology that remembers without burning, that thinks with almost no motion at all, China is quietly proposing a different temperament for the machines to come: less like a roaring engine, more like a stone in a river—silent, patient, shaping the flow with almost invisible effort.
FAQ
What exactly is the “resurrected” technology?
It refers to families of analog and analog‑like memory and computing approaches first explored around the 1960s–1970s—such as optical or in‑memory computing—that store information as stable physical patterns rather than as constantly refreshed digital bits, now updated with modern materials and AI concepts.
How can it use up to 200 times less energy than digital chips?
Because these devices don’t need continuous power to maintain their state and can perform certain computations as a natural result of how signals pass through their structure, they avoid much of the repetitive switching and data movement that dominate power use in conventional digital systems.
Will this replace existing digital computers?
Unlikely. It’s better suited as a complement, handling specific, pattern‑heavy tasks (like local AI inference or sensing) while digital processors continue to manage general‑purpose, flexible computing.
What are some realistic applications?
Edge AI in cameras and sensors, low‑power environmental monitoring, industrial inspection, smart infrastructure, and any scenario where small, task‑specific intelligence must operate for long periods on tiny energy budgets.
Is this technology available commercially yet?
Most work is still in the research and prototype phase. Some early analog and in‑memory computing chips are emerging, but widespread commercial products using the full potential of these resurrected ideas are likely a few years away.
Why is China particularly involved in this revival?
China faces strong incentives to reduce the energy footprint of its growing AI and cloud sectors and to diversify beyond conventional semiconductor paths. Reviving and modernizing these older analog concepts offers both potential energy savings and a distinctive avenue for technological leadership.
What are the main challenges ahead?
Ensuring device reliability and longevity, scaling manufacturing, integrating with existing digital ecosystems, managing analog noise and variability, and developing programming tools that make it easy for engineers to design useful systems with these unconventional architectures.