Today, you probably asked a question of a large language model, or accepted a connection suggestion on LinkedIn, or watched a recommended video on YouTube, or took a different route to work based on a traffic prediction from Google Maps. In other words, you probably used artificial intelligence. But what you might not know is how much energy that interaction consumed or why.
AI requires processing massive amounts of data, which is usually done in large data centers populated by thousands of GPUs capable of executing up to trillions of operations per second. But each of those GPUs achieves that by consuming as much as 1,000 watts apiece. For comparison, if you’ve got a newer smartphone, it probably uses less than 1 W. That kilowatt figure puts GPUs on the same level as vacuum cleaners, dishwashers, and stoves, but with the big difference that data-center processors are operating uninterrupted around the clock.
Fundamentally, a lot of this inefficiency is because GPUs are trying to simulate the workings of artificial neural networks using software and billions of transistors, which requires using energy to move massive amounts of data. What’s more, the simulated artificial neurons that make up these networks lack even a fraction of the complex computing behavior of the biological neurons that comprise the most energy-efficient computing system that we know, the human brain.
Huge amounts of work have gone into making electronics operate more like biological neurons and synapses. Some research has focused on developing new, experimental devices, but they aren’t yet reliable enough to be used in large systems. Other efforts aim to implement neurons and synapses by interconnecting many complementary metal-oxide-semiconductor (CMOS) transistors—the workhorses of digital logic—to simulate a single neuron and synapse. But this approach requires so many transistors (and a few bulky capacitors) that it greatly limits the size of the system that can be constructed, making it unclear how such brain-inspired hardware could ever scale up and compete with state-of-the-art GPUs.
But all along there was an artificial neuron and a synapse—each a single device—hiding in plain sight. We found them last year. They were each made possible by an ordinary CMOS transistor—and not even a very good one at that. This is the story of their accidental discovery and their great promise for lowering the environmental footprint of AI.
Biological and artificial neurons
Modern digital electronics is based on producing and manipulating the ones and zeros of the binary code through the operation of metal-oxide-semiconductor field-effect transistors. MOSFETs have evolved in recent years, but their classic form consists of a piece of silicon that has been doped to contain an excess of either positive (p-type) or negative (n-type) charge carriers. (CMOS logic contains transistors of both types.) The device has two terminals connected to the silicon through regions highly doped with the opposite polarity of the rest of the silicon—the source and the drain. Another terminal, the gate, sits atop the silicon that separates the source from the drain. The gate itself doesn’t connect directly to this silicon, instead resting above a thin layer of insulating dielectric.
Advertisement
Notably, there is a fourth terminal that attaches to the bulk of the silicon; think of this bulk terminal as connecting to the underside of the chip. It doesn’t typically get much attention, but it’s very important to our story.
When voltage is applied at the gate and the bulk terminal is grounded, charge carriers of the same polarity as the source and drain are attracted to the channel region. In the case of an n-type source and drain, that will be electrons; for p-type it will be holes. The presence of these charges forms a conductive channel that reduces the resistance between the source and the drain by several orders of magnitude, and the device switches on. As the voltage at the gate increases, this physical phenomenon produces a current signal that, when plotted against the gate voltage, rises steadily. This response is ideal for logic gates, converters, multiplexers, memories, and other digital circuits. But it is not a good fit for mimicking the behavior of a neuron.
In real neural tissue, brain cells, called neurons, consist of a cell body, a long projection called an axon, and short branching projections called dendrites. The suite of behaviors and computing this collection of components is capable of is rich and broad, but the portion that artificial neural networks hope to copy is this: When the cell body’s voltage is perturbed enough to reach a particular threshold, a self-propagating pulse of voltage, called an action potential, shoots down the axon. The axon terminates in a synapse, an electrochemical connection between the axon and another neuron’s dendrites. The action potential will then temporarily boost the voltage of this next neuron, by an amount that depends on the strength of the synaptic connection. If enough action potentials reach these dendrites in a given time—from this neuron or from others that might also form synapses there—the cell body’s voltage will surpass the threshold and trigger its own action potential.
To get closer to the behavior of real neurons, artificial neurons should produce a current spike when a critical voltage threshold is crossed and then quickly relax back to a resting state on their own. This spike needs to be sudden—nonlinear. It should also exhibit some hysteresis; that is, the activation and relaxation voltages should be different from each other to ensure that current flows only for a certain amount of time.
Advertisement
What’s wanted from an artificial synapse, the thing that connects two artificial neurons, is less complicated, but equally important. The main thing is that its conductance can be electronically adjustable. The device’s conductive states should increase and decrease in a linear pattern and remain stable over time.
No single MOSFET working under the standard operation mechanism can reproduce either of these neural properties. Instead, it’s been done by combining them into complex circuits. Until now, each neuron and each synapse has been implemented by interconnecting dozens and sometimes even hundreds of MOSFETs, which is highly inefficient in terms of area, performance, and cost. To limit the amount of space needed, chips can multiplex their signals, sending them to neurons and synapses serially, but such sequential processing introduces additional delays.
Despite these area-and-time penalties on tasks such as audio processing, computer vision, or health monitoring, state-of-the-art brain-inspired microchips have achieved power reductions up to a thousandfold compared with those of GPUs or CPUs on the same task. If we could create neurons and synapses from individual devices that are readily manufacturable instead, we might target more massive implementations while maintaining energy efficiency.
Reinventing the MOSFET for AI
Working in our laboratory in 2024, one of my students was measuring a memory circuit that consisted of one transistor and one memristor—a type of nonvolatile memory device first fabricated in 2008. The student’s memristor circuit was built from two-dimensional material atop a silicon microchip containing MOSFETs. The MOSFETs were created in a commercial foundry using fabrication technology called the 180-nanometer node, which was cutting-edge in the year 2000.
Advertisement
One day the student forgot to connect the bulk terminal of the transistor. What he observed was a sudden increase in current with high nonlinearity that self-relaxed when the voltage was ramped down (a phenomenon called a hysteresis loop). This was a very promising neuronlike behavior!
After a fruitless week of trying to think of an explanation for this behavior, I (Lanza) asked Pazos, then my postdoctoral fellow, to try to observe and control this phenomenon in chips without memristors. This time, we applied pulses of voltage—like the spikes a neuron would produce—instead of the ramped voltage that my student used when he first saw the peculiar behavior.
Pazos’s new data helped us understand what was going on. The key was that oft-ignored fourth, or bulk, terminal of a MOSFET. Under ordinary operation, many mobile charge carriers flitting through the channel collide with the silicon atoms, producing free pairs of electrons and holes—a process known as impact ionization. The electric field created by the potential difference between the source and the drain causes these new free electrons to drift toward the positively biased drain and the holes to move toward the bulk terminal, which is usually grounded, removing the charge without any drama.
However, when the bulk terminal of the transistor is floating—unconnected as it was in my student’s experiment—the holes produced by impact ionization cannot be driven to the ground. Instead, they accumulate in the bulk of the silicon, increasing its voltage. Then things start to get interesting.
Advertisement
It helps here to imagine a MOSFET as two different kinds of transistors occupying the same physical space—the intentionally constructed MOSFET and a hidden, bipolar junction transistor. A bipolar device transmits a current signal across two p–n junctions, in this case the interfaces between the source and the channel region and the channel and the drain. This signal is in proportion to a smaller current at a third terminal in between, called the base. In our experiment, that third terminal is the bulk.
To get current flowing through a bipolar transistor, you need a big enough potential difference between the base and one of the other terminals, so that current can get across the p–n junction. Let’s say this “threshold voltage” is 0.7 volts, although the real number depends on device geometry and silicon doping. In our device, that potential difference comes from those holes that were accumulating in the bulk, because it was not connected to ground. Once it reaches the threshold voltage, the device becomes sharply conductive, producing an abrupt increase of current. This sharp current increase eventually falls off once the drain voltage is lowered, because that lowering reduces the rate at which holes are generated in the bulk. The remaining excess holes recombine with stray electrons or leak away, and finally the bulk voltage falls. This cycle of hole accumulation, current spike, and hole removal gives rise to a hysteresis loop, very much like the electrical behavior of a biological neuron as it integrates ionic currents, fires a spike, and relaxes back to its resting voltage.
Initially, we observed this behavior only in a few transistors, and the relaxation time was very different for each of them. So, to try to control it better, we adjusted the resistance of the bulk terminal using a second MOSFET. Simply setting that resistance suddenly caused all the transistors to fire at the same voltage with hardly any variability. In other words, we found we could create perfect electronic neuron behavior in a single silicon transistor by controlling the bulk contact resistance. Setting the resistance can be done by doping the silicon during fabrication, but we think the two-transistor cell—where one acts as the bulk resistance—offers much greater versatility because it allows for electronic control.
We had to make sure the phenomenon would last, otherwise such a device would be useless. To our delight, every single one of the devices we tested worked over 10 million cycles. Not even one of them failed during our tests.
Advertisement
To be honest, we were amazed. Dozens of research groups and companies all around the world have spent many millions of U.S. dollars over the past 20 years trying to emulate these neural behaviors using experimental memristor-like devices and other things, with limited success, mainly due to reliability and cost issues. We managed it in the cheapest and most industry-standard device: the MOSFET. This result was so shocking that we decided to confirm it using microchips from a different foundry. It was successful: All the behaviors could be reproduced, and perfect yield was achieved once again.
We were happy with the results and had started the process of filing for a patent and writing up our findings for the journal Nature, when our lab made another astonishing discovery: The same kind of MOSFET could act as a synapse, too!
Recall that in ordinary operation some electrons crash into silicon atoms to create pairs of electrons and holes. We noticed that at specific values of bulk resistance a significant amount of the charge from this impact ionization would get trapped in the gate dielectric. This trapped charge interferes with the flow of current through the MOSFET, effectively changing the device’s conductance. Importantly, this new conductance is stable and adjustable at will. It was then that we realized the MOSFET could also be used as an electronic synapse.
As it was in the neuron transistor, the bulk terminal was the key. A negative bulk-source voltage drives electrons into the dielectric, decreasing conductance. A positive one pushes holes in, increasing it.
Advertisement
From neuromorphic device to circuit to system
Here’s how the MOSFET synapse and the MOSFET neuron, together called a neurosynaptic random-access memory, or NSRAM, could work together to achieve a simple neural circuit: Say you had a circuit consisting of three synapse MOSFETs and a neuron MOSFET. The synapses have already been programmed as we’ve described, so that each has a different conductance. Spikes of voltage with different patterns and frequencies are applied to the gate of each of these transistors. What emerges from their drains are spikes of current with amplitudes modulated by the synapses conductance values.
The spikes converge at the drain of the neuron MOSFET. With each spike, impact ionization causes charge to build in the bulk of the silicon. Some of it will drain away, but if enough spikes arrive in a short enough period of time, the bulk voltage will reach a value at which the “hidden” transistor triggers a spike of current through the MOSFET. This current would then go on to become the input to other MOSFET synapses, and so on. The behavior is exactly the kind of integrate-and-fire action real neural circuits deliver.
The competitive advantage of our single-MOSFET electronic neurons and synapses is straightforward: We can produce with only one or two transistors the electronic signals that today require, at an industrial level, dozens and sometimes even hundreds of components. And moreover, unlike other emerging technologies, our solution is fully compatible with today’s silicon manufacturing lines and exhibits a yield of 100 percent in key figures of merit with near-zero variability.
Building functional circuits for brain-inspired computing and AI based on this technology is as exciting as it is laborious. It will require us to improve our computer models to resemble the behavior of both devices more accurately and to do so with computational efficiency. We must also perform accurate circuit- and system-level simulations to validate computing architectures, design peripheral circuitry to drive and convert signals, and undergo multiple fabrication rounds to optimize performance.
Advertisement
But all that will be worthwhile, because it could result in brain-inspired microchips for AI with better energy efficiencies than what we have now. These chips will first be a fit for smaller-scale, “edge-AI” tasks, such as bringing greater intelligence to battery-powered systems. But if we can scale up such chips, maybe in the long run they can compete with state-of-the-art GPUs.
ISBNdb ships up to a million books anonymously to AI labs
Pre-2022 books are prized because chatbots never touched their text
Spine-cutting scanners destroy originals to speed up digitization for training
AI companies are increasingly turning to printed books published before 2022 as preferred training material because those works predate the widespread use of AI-generated content.
Large-scale scanning operations reportedly involve cutting book spines, separating pages, and destroying physical copies to create digital datasets for large language models.
The practice has attracted growing criticism because some books entering these pipelines are reportedly extremely rare, raising concerns about irreversible cultural losses.
Latest Videos FromTechRadar
Advertisement
Pre-2022 books become valuable AI training material
Reports by 404 Media found data broker ISBNdb supplies physical books in bulk to AI developers seeking human-written material unaffected by modern chatbot output.
The company argues books published before 2022 offer cleaner datasets because they cannot contain text generated by contemporary large language models.
They are often considered “dense, edited, authoritative,” in contrast to internet content increasingly filled with machine-generated material of uncertain quality.
The approach also attempts to avoid so-called model collapse, in which AI systems gradually lose quality after repeatedly training on synthetic content generated by earlier models.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
ISBNdb additionally argues that older printed works avoid deliberate data-poisoning techniques authors increasingly use to disrupt AI training through carefully modified documents.
Advertisement
However, there are reports that many of these books are scanned using high-speed equipment.
This equipment requires workers to remove the spine before feeding individual pages through automated imaging machines.
That process reportedly destroys the original volume, making rapid digitisation considerably cheaper than slower preservation methods designed to keep books physically intact.
Advertisement
Secrecy and legal rulings fuel preservation concerns
ISBNdb openly acknowledges reputational concerns surrounding the practice while offering strict non-disclosure agreements that keep customer identities confidential throughout commercial engagements.
Its website reportedly states, “‘AI company destroys two million books’ is not a headline that generates sympathy,” while suggesting clients describe the process as digital preservation.
Such a level of destruction is an order of magnitude bigger than the loss of the Library of Alexandria. Yet, it is unfolding with none of the outrage that history reserves for burned libraries.
Advertisement
Booksellers interviewed by 404 Media said some volumes entering these scanning programmes have very few surviving copies after enduring wars, fires, and centuries of handling.
Critics argue that unlike websites or widely available modern publications, exceptionally scarce historical works cannot simply be reproduced after their physical copies disappear forever.
A recent United States court ruling involving Anthropic found that scanning legally purchased books for AI training constituted fair use under specific circumstances.
Part of that reasoning held that destroying each printed copy during scanning meant one legal copy effectively replaced another rather than creating multiple copies.
Advertisement
In response to a critic (@Hedgie) of this method on X, Elon Musk said, “I’ve asked the SpaceXAI team to preserve any rare books in a library and scan them the hard way,” suggesting an alternative approach.
If significant awareness is not created, this quiet erasure of irreplaceable books risks becoming the defining act of cultural loss for this era, remembered only after it can no longer be undone.
For more than four decades, the U.S. Navy’s Ticonderoga-class guided missile cruisers have served as some of the fleet’s most capable surface combatants. Designed around the powerful Aegis Combat System, they have played a central role in protecting aircraft carriers, defending against enemy aircraft and missiles, and coordinating complex naval operations around the globe. However, after years of service, the Navy has begun retiring the aging class, with more ships scheduled to leave service over the remainder of the decade.
The challenge is that the Navy has no direct replacement waiting in the wings. Unlike previous generations of warships that were succeeded by clearly defined new classes, the retirement of the Ticonderoga-class vessel leaves a capability gap the service is still working to address. While several options are being considered including relying more on destroyers, no single platform is expected to immediately assume every mission currently performed by these cruisers. That uncertainty has made the future of the Navy’s large surface combatant fleet one of the most closely watched questions in modern naval planning.
Advertisement
Why the Ticonderoga fleet is disappearing
The Navy originally built 27 Ticonderoga-class cruisers beginning in 1983 at a cost of roughly $1 billion each. Over time, however, age has taken its toll. 20 of these ships have already been decommissioned while the remaining seven vessels are expected to retire by 2030 as maintenance costs continue to rise and the ships become increasingly expensive to modernize. Although several cruisers have received temporary service-life extensions, Navy officials have made clear that the class is approaching the end of its operational life.
Advertisement
That doesn’t mean the Ticonderoga-class cruiser has become ineffective. On the contrary, they remain among the Navy’s most capable air-defense platforms. The problem is that keeping 40-year-old warships mission-ready has become increasingly difficult and costly, particularly as newer technologies continue to emerge. Earlier efforts to modernize portions of the fleet ultimately proved unable to keep every cruiser in service. As a result, retirement has become less about combat capability and more about balancing readiness, maintenance demands, and long-term fleet modernization.
Advertisement
What could replace the Ticonderoga-class?
For now, the Navy’s most practical solution is not another cruiser at all. Instead, many of the Ticonderoga’s responsibilities are expected to transition to the latest Flight III Arleigh Burke-class ships such as the USS Ted Stevens sporting the DDG (guided missile destroyer) classification. Equipped with the advanced SPY-6 radar and the latest version of the Aegis Combat System, these new destroyers can assume many of the fleet air-defense and command responsibilities traditionally handled by the retiring cruisers. While they cannot replicate every capability of the Ticonderoga class, they provide the Navy with an effective bridge while longer-term plans continue to develop.
Looking further ahead, the Navy expects its future DDG(X) program to eventually become the next generation of large surface combatants. However, that program remains years away, with the first ships not expected until the early 2030s. Until then, the Navy finds itself in the unusual position of retiring one of its most successful classes of warships without a direct one-for-one successor ready to take its place. Upgraded destroyers may help fill much of the gap, but exactly what ultimately replaces the Ticonderoga-class remains an open question — one that will shape the future of the U.S. surface fleet for years to come.
Ordinary glass stops electricity cold. It lets light through and blocks current completely. That combination works fine for windows and phone screens, yet it leaves the material useless as any kind of switch. A thin coating of indium tin oxide changes the rules. The layer is only 10 to 300 nanometers thick, nearly invisible, and still conductive enough to carry current from one edge of a small pane to the other.
Sokol ordered a pack of this coated glass for another project and decided to use a couple of the sheets as buttons. He took his multimeter to test the coating and was thrilled to see that it performed as predicted. One side of each sheet measured around 20 ohms, but the uncoated side remained nice and open, as you would expect from plain glass. The conductive coating is only applied to one side of the glass, so bear this in mind when using it.
Flip Into Play – Flip down, ready to play. Take a call or reply to a message mid-game—just flip it open without unplugging. The hinge is rated for…
Vertical Mobile Gaming – Designed around portrait play, FlipPad brings a classic handheld feel to your phone. Perfect for retro games and…
USB-C Plug and Play – Built-in USB-C connection makes setup simple and direct. Plug in and start playing instantly. The USB-C port is rated for over…
The button is made up of two of these panes, with their coated sides facing each other and a thin sheet of paper between them as a spacer. One pane is grounded, while the other is connected to an Arduino pin equipped with a pull-up resistor that keeps the pin at 5 volts even when the circuit is open. As long as the two conductive layers are apart, the pin reads high and the built-in LED remains dark; however, press down on the stack, and the glass flexes just enough for the coatings to make contact. Then current begins to flow, the pin drops to zero, and the LED illuminates immediately.
That’s the entire system, so no fancy sensors or drivers, just the old ITO film. The paper spacer keeps the surfaces from touching until you exert pressure, at which point the glass springs back and breaks contact when you let go. The whole thing is a little rudimentary and rough around the edges, but it does the job.
The same two sheets can do more than simply turn on and off. You can use the same concept to create a resistive touchscreen by applying a voltage gradient to one of the coatings, with half at 5 volts and half at zero. When the other coating makes contact, it detects the voltage where it touches and transmits it to an analog pin, which is where you receive your readout indicating where the push occurred. You can add another gradient to the opposite layer, resulting in two-dimensional placement. Sokol points out that a more polished version of this would require copper tape for consistent edges, external resistors for stable gradients, and larger glass, but the core concept is already present in the simple button. [Source]
I’ve been a fan of PSB Speakers since I first auditioned the brand’s M4U 2 headphones back in 2013. Over its 50-plus years of operation, the Canadian outfit has consistently shown a knack for squeezing big sound into small packages. Never was that more apparent than in 2023’s Alpha iQ, a surprisingly versatile pair of powered bookshelves that pull more dynamism and musical expression out of a four-inch woofer and micro-tweeter than they have any right to.
The follow-up iQ1 and iQ2 stand firmly on the Alpha’s shoulders, offering similarly potent musicality and connectivity packed into miniature cabinets, with a woofer-tweeter stack swap that always throws my brain for a loop. The new pairs also share the Alpha’s flair for stylish minimalism, especially the pricier iQ2, which adds upgrades such as a wireless connection between the speakers and a more refined veneer in colors like Ember Red and Sandstone Beige.
PSB Speakers iQ2 Colors
While the iQ’s hardware is all PSB, daily operation relies heavily on BluOS’s powerful software, which handles virtually all controls and audio settings, provides access to more than 20 streaming services, and even supports multiroom audio with rock-solid stability. Unfortunately, I can’t say the same for the iQ2’s HDMI eARC TV connection, which proved doggedly inconsistent throughout my review. The iQ2 worked great for other use cases, but without a stable way to connect seamlessly to a TV in today’s market, the package falls well short of its promise.
Related Reviews:
Setup: Long Haul Connection
PSB Speakers iQ2 (front)
Setting up the iQ2 isn’t a major hassle for tech-savvy users, but it’s not the push-button start you’ll get from multi-room setups like Sonos, recent app issues notwithstanding. PSB’s quick guide lays out a three-step process that includes simply plugging in the speakers and downloading the BluOS app for Wi-Fi setup, but the online manual hints that Ethernet is the best option where available. As if that were an omen, the iQ2 balked at my network’s 5 GHz Wi-Fi band during an AirPlay connection, leading to multiple stallouts.
Swapping to my split-off 2.4 GHz band fixed the issue, prompting an update and a paint-drying 20 minutes or so of total connection time between the speakers. The process included blasting surprisingly loud music from the primary speaker, which houses the connection hub, to designate it as the right or left channel, but I appreciate the ability to swap the orientation as needed. Users with particularly troublesome rooms should note that the iQ2 doesn’t offer dedicated calibration.
Advertisement
Controls and Usability: HDMI Issues
PSB Speakers iQ2 (rear)
There’s also a learning curve for those new to BluOS, or even those like me who’ve been away for a few years. For example, going into the Players tab to find the iQ2’s settings unlocks basic playback controls and options like adding a subwoofer, but there’s no way to select an input. For that, you’ll need to consult the Music tab, which includes all physical inputs, input settings such as allowing the speakers to switch automatically when a signal is detected, and access to BluOS’s 20-plus streaming services and internet radio.
It feels overengineered for a pair of speakers, but that’s because the BluOS ecosystem is designed to let you group the iQ2 with more than 60 other compatible players throughout your home from brands like NAD, DALI, and Bluesound. The app is stable and thoughtful, with features such as haptic volume control, but I wish PSB included a physical remote for quicker access to inputs and other settings. Call me old-fashioned. You do get a few controls on top of the primary speaker, including play/pause, volume, and two programmable presets.
Controls aside, the iQ2’s hardware troubles were the real frustration. HDMI eARC is designed to provide a seamless, high-resolution connection between your TV and speakers, but the iQ2’s implementation was a mess. I experienced audio cutouts across multiple TVs, including instances where the TV recognized the speakers as a source but played no sound. PSB Speakers’ PR team told me this is a known issue and pointed to a firmware fix, but it didn’t work for my review pair.
Design & Features: Elegant Minimalism
PSB Speakers iQ2
The iQ2 make a great first impression, especially in the Sandstone color I reviewed, looking both stylish and understated. The cabinets feel well built, and their fingerprint-resistant veneers ensure that the refined aesthetic holds up. There’s solid heft to the compact cubes, though at under 10 inches tall, they looked almost comically small against the 75-inch Micro RGB TV I first paired them with, especially with their little magnetic hats that serve as acoustic grilles.
That said, the iQ2’s wireless connection means they can easily adapt to larger screen sizes, and there’s nothing comical about their 270 watts of Class D amplification. That works out to 90 watts for each speaker’s four-inch polypropylene mid-bass driver and 45 watts for each 0.75-inch aluminum-dome tweeter. PSB claims a frequency response of 65 Hz to 20 kHz, and the speakers dive toward that lower number with more authority than their size suggests.
A column of inputs on the back of the Primary speaker includes HDMI eARC, optical, USB-C playback, USB-A for external drives, a dual line-in/MM phono RCA input adjustable in the app, Ethernet, and a subwoofer output.
Advertisement
Advertisement. Scroll to continue reading.
Streaming support includes Spotify Connect, TIDAL Connect, Qobuz Connect, and Apple AirPlay 2. You’ll find many more options in the BluOS app, including Deezer, Napster, and iHeartRadio, to name a few, but not Apple Music. Other streaming options include access to internet radio stations, local network files at up to 24-bit/192 kHz, and Bluetooth with aptX Adaptive.
Listening: Mighty Minis
PSB Speakers iQ2
I started testing the iQ2 via lossless Spotify Connect and was quickly reminded why PSB holds a special place in my mind’s cluttered library of audio brands. A bit like Canadians themselves, the iQ2 are unassuming at first, but digging deeper reveals an effective mix of confidence and depth. Their sound signature is clean, smooth, and effortlessly natural, with accurate stereo imaging and more bass than you’d expect from micro speakers.
With the iQ2 playing in the background as my pre-toddler arrived home from daycare, I was constantly distracted by their penchant for drawing out instrumental timbres, from satiny brass and woodwinds to gritty electric guitars and poppy percussion. They do a good job of leaning into different genres, with enough finesse to give each instrument its own space, from the dual vocals in Too Short’s “Money in the Ghetto” to the cacophonous horns in The Beatles’ “Good Morning.”
The warm midrange can make more laid-back mixes sound a little flat, but its keener topside adds some excitement to higher-register instruments and effects without feeling sharp or brittle. The sweet yet pointed treble makes easy work of speedy transients like the reverberating synths in Sweet Spirit’s “Baby When I Close My Eyes,” fluttering the effect back and forth across the stereo image with a deft touch that stands out more readily than in options like SVS’s Prime Wireless Pro.
Advertisement
Bass is firm and well-rounded, if not particularly revelatory, until you remember the size of the drivers, which outflex just about any four-inch woofers I’ve encountered, including options like U-Turn’s more rugged-sounding Ethos speakers. Even some larger 5.25-inch drivers provide less mid-bass punch. It’s enough groove to be tantalizing, but if you’re after raw emotional power, you’ll need a subwoofer. As usual, my SVS 3000 Micro was a great pairing, firming up the lowest frequencies with richer musicality and some sledgehammer punch while opening more space in the upper registers.
Turning to vinyl with my Orbit Theory turntable, the speakers showed off solid detail and dynamics on my go-to albums, such as Brubeck’s Take Five. Paul Desmond’s buzzy sax in “Blue Rondo” was rendered with keen precision, pulling out most of his breathy overtones. Joe Morello’s stick work had an expressive touch in moments such as his multitone tom play during the title song’s drum solo. I was less impressed when moving to the iQ2’s internal MM phono stage, which was both quieter and less precise than the Theory’s internal preamp, though that seems to be a common issue with powered speakers in this class. You can often do better with an external preamp.
Moving to TV and films, when the HDMI eARC port was working, the iQ2 proved more than up to the task of providing exciting sound effects at the sides of the stereo image and clear dialogue at the center, elevating the subtle diction in well-mixed productions like House of the Dragon. Again, the speakers brought a natural liveliness to effects such as stone hallways and bridges, crackling fires, and the woody creak of the show’s many oversized doors.
PSB Speakers iQ1
The Bottom Line
On paper, the PSB iQ models, especially the iQ1, provide a ton of value. You’ll spend a lot more to get the pricier iQ2’s wireless connection and fancier cabinets, but both models offer impressive connectivity and features matched by surprisingly good performance, especially in smaller rooms.
Advertisement
If not for the HDMI eARC issues I encountered, either pair would be a top pick for those after a compact system with connectivity to spare, not to mention the ability to group it with other systems in the BluOS family. As it stands, I can’t recommend the iQ1 or iQ2 until PSB Speakers can ensure simple, reliable connections for all your components.
For those seeking alternatives to the cheaper iQ1, I recommend SVS’s steadfast Prime Wireless Pro, which offers well-rounded sound and great connectivity in sleek, piano-gloss cabinets. If your budget is more in line with the iQ2, KEF’s LSX II provides similarly stylish design and colorways, along with even better sound performance. Neither pair offers multiroom capabilities, however, so if that’s important, you may need to look elsewhere or wait for PSB to provide a real fix.
Pros:
Smooth and natural tonal balance
Good instrumental separation
Powerful bass for their size
Stylish, well-built cabinets in multiple colors
Loads of connectivity options
BluOS multi-room audio support
Cons
Setup and basic control could be simpler
HDMI eARC cutouts are a mess
Upgraded cabinets and wireless connection are pricey
The AI boom made GPUs famous. The dull chips that wire thousands of those GPUs together are quietly becoming just as valuable.
Xsight Labs, an Israeli chipmaker, has raised more than $300 million at a $2.8 billion valuation, it said on Thursday. Fidelity led the round. That is more than five times what the company was worth in 2021, a jump it made in a single year.
Xsight builds the plumbing of the AI data centre. Its two chips, the E1 processor and the X2 switch, shuttle data between servers, memory and networks. As models grow, moving that data has become one of the industry’s tightest bottlenecks, and the wiring between GPUs is now a prize of its own.
Taking on Nvidia’s other monopoly
Nvidia does not just sell the GPUs. It also dominates the networking layer, which it bought its way into with Mellanox. Xsight’s pitch is the opposite of that closed kit.
Advertisement
Its chips are open and programmable. They run standard Linux and SONiC, and operators can rewrite the X2 switch in software after it ships, instead of waiting for new silicon. The market wanted “a high-performance alternative to closed, legacy architectures,” said chief executive Yossi Meyouhas. Xsight joins a wave of startups chipping at Nvidia’s grip on the AI stack.
Already in orbit
This is not a lab project. Several global network operators have picked Xsight’s silicon, including SpaceX’s Starlink. Last week its processors passed a test aboard SpaceX’s new Starlink V3 satellites, Globes reported. Programmable chips suit space, where engineers fix a fault in software rather than swap out the hardware.
The selling point is efficiency. The E1 is the first 800-gigabit data processor to ship, and runs under 75 watts. The X2 moves 12.8 terabits a second while drawing 40% less power than older switches. As AI’s electricity bill balloons, every watt saved is a pitch.
A serial exit machine behind it
The pedigree helps. Veterans of EZchip, the Israeli networking firm Mellanox bought in 2016, founded Xsight in 2017, CTech reported. Its chairman is Avigdor Willenz, a serial chip entrepreneur who has sold past companies to Marvell, Amazon and Intel.
Advertisement
His backers followed him in. The round drew Fidelity, Intel Capital, T. Rowe Price, Battery Ventures and Valor Equity Partners, among others. The cash will fund the next chip generations and a bigger sales push at Tier-1 hyperscalers.
The money pouring into AI infrastructure is no longer chasing only GPUs. It is chasing everything around them, from power stations to the wires in the rack. Xsight is betting the winners in networking will be the open ones.
This week on the GeekWire podcast: Microsoft and Amazon both reported quarterly numbers, and both stocks rose on cloud results that beat expectations. Is all that AI spending paying off? And in related news, Microsoft sees a rare annual headcount decline, hitting product R&D hardest.
Plus: Satya Nadella builds a Power BI dashboard out of an analyst’s research report, and touts it on the earnings call to make a bigger point. Jeff Bezos names Amazon’s chips business as the long-awaited fourth pillar. And AI House managing director Jacob Colker delivers a much-needed pep talk for Seattle tech, calling on the region to recognize and build on its strengths.
Got an idea and want to make a simple 3D model, but don’t want to install a full-fledged CAD modeling suite and queue up a few hours of tutorial videos? Check out SketchForge, a 3D modeling program that runs locally in one’s browser without any need for an account, or external services.
SketchForge takes a more WYSIWYG approach to 3D modeling by making it easy to put an object together with primitive shapes, and making it extra easy to specify dimensions and align parts with one another. There’s also a sketch feature that makes it easier to create more complex shapes by making a 2D drawing, then extruding or revolving it into a solid. We like that it has STEP format export as an option, making it easy to import your creation into another CAD program of your choice later. Most 3D printer slicers natively support the STEP format nowadays, too.
It’s a bit reminiscent of Tinkercad in concept, but entirely local. It’s still new, but there’s a demo online that gives a good idea of its capabilities if you’d like to give it a spin.
Advertisement
We’re reminded of CaDoodle, another project that takes the “Tinkercad, but local” approach but as a standalone executable, instead of browser-based.
OpenAI CEO Sam Altman seemed excited to share what he called a “cool use case” on Friday, posting that parents could “connect your family calendars and explain your kids’ interests” to the company’s new product ChatGPT Work, then “every morning for the drive to school, have it make a podcast that talks about one kid’s soccer game that afternoon, one kid’s upcoming birthday, some news, etc.”
Altman’s post prompted lots of virtual eye-rolling, including a simple response from Alex Hirsch, creator of the animated series “Gravity Falls”: “What if you just talked to your children”?
Hirsch’s question seems to have gone even more viral than Altman’s initial post — while Altman’s comment has been reposted around 300 times and liked around 9,600 times (as of Saturday morning), Hirsch’s response has been reposted 9,000 times and received 122,000 likes.
This isn’t the first time a tech CEO has promised that AI can shield users — and specifically users’ morning commutes — from the messy realities of human experience. Last year, Microsoft CEO Satya Nadella said he’d stopped listening his favorite podcasts during his morning drive; instead, he asked an AI chatbot about those podcasts.
Advertisement
Altman’s post also echoed comments he’s made about parenting in the past, most notably when he appeared on “The Tonight Show with Jimmy Fallon” and declared, “I cannot imagine having gone through figuring out how to raise a newborn without ChatGPT.” (Though he was also quick to acknowledge, “Clearly, people did it for a long time, no problem.”)
Getting parents on-board seems to be a priority for OpenAI, which recently posted a job listing for a product manager with experience building trust-sensitive consumer experiences for parents and families.
But while the company has added safety features for parents, it also faces multiple lawsuits from parents and families alleging that ChatGPT played a role in loved ones’ delusions and suicides. (The company has said it’s “continuously improving how our models respond in sensitive interactions.”)
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Back in the day, you could just scroll the App Store, pick out an icon or two that you liked the look of, and have a new game downloaded and ready to play in just a few moments.
Ultimate Guide to Apple Arcade
(Image credit: Apple / Future)
This article is part of our Ultimate Guide to Apple Arcade: the definitive guide to everything that you need to know about the gaming service. Be sure to browse the full series here.
Advertisement
Booting them up, you’re not confronted with a 29.68GB additional download that you have to stay on one screen to complete, a wall of sketchy in-app permissions to accept, or a page demanding that you sign in with some iffy third-party account — just a neat little self-contained experience designed to help you kill a few hours on the train.
Though they seemed simple, many of the best App Store games were quietly revolutionary. Angry Birds and Fruit Ninja pioneered intuitive touch-screen controls, while Doodle Jump smartly utilized the iPhone’s accelerometer to show how hardware features could become core gameplay mechanics.
Statista estimates that there are now more than 180,000 mobile games available on the App Store. But when was the last time you installed one that felt genuinely exciting, rather than like a thinly veiled scheme to force ad views, harvest personal data, or pressure you into pricey Gacha Pulls? That’s exactly where Apple Arcade comes in.
Advertisement
Back in time
Puyo Puyo Puzzle Pop is an incredible Apple Arcade exclusive entry in the long-running puzzle series. (Image credit: Sega)
Having spent the last week diving into all 200+ of the titles available on the service, I can’t help but think that Apple’s highly curated library of mobile titles feels like a journey back in time to the App Store’s glory days.
The whole experience is completely seamless. You sign up for $6.99 / £6.99 / AU$9.99 a month and can immediately jump into browsing the games. They’re all contained in a dedicated Arcade tab in your iPhone’s App Store, so there’s no ambiguity about whether the app you’re looking at is part of the service or not.
Advertisement
Sign up for breaking news, reviews, opinion, top tech deals, and more.
Although the selection here is nowhere near as expansive as the regular games tab, it feels genuinely curated and cuts out huge amounts of clutter and spam.
No matter what you like to play, there’ll be at least a few games for you. An impressive variety of genres are represented in the catalog: there are plenty of massive role-playing games (RPGs) that you can sink hours into, be that Trials of Mana+ or Final Fantasy+, fast-paced racers such as Asphalt 8: Airborne+ and Sonic Racing, head-scratching puzzlers in Monument Valley+ and Lego Builder’s Journey, and so much more.
Jetpack Joyride 2 adds new missions, weapons, upgrades, and more. (Image credit: Halfbrick Studios)
Those iconic early mobile titles I mentioned earlier? They’re back too in new and improved forms that make them a nostalgic joy to revisit. There’s a full-on Angry Birds remake with enhanced graphics and extra levels, an expanded Doodle Jump sequel that updates the hand-drawn graphics and adds new environments, and a Temple Run successor that trades the endless format for challenging levels that are better-suited for short bursts of play.
Advertisement
The innovation of the early App Store is there too. RPG Fantasian wows with incredible art direction that sees you exploring hundreds of beautiful real-world dioramas that were digitally scanned into the game, while bonkers sports simulator What the Golf? takes full advantage of punchy iPhone haptics to add real weight to its wacky physics challenges.
No distractions
No ads makes it easier than even to lock-in in rhythm games like Taiko no Tatsujin Pop Tap Beat. (Image credit: Bandai Namco)
The games are all entirely devoid of ads and microtransactions, so you never have to deal with a frustrating splash screen or pop-up that takes you out of the action.
Advertisement
This means that most of the games will work entirely offline as well, just like they did back in the day. I had practically given up playing mobile games on my commute before getting into Apple Arcade simply because it was so hard to find titles that wouldn’t kick me out the second my train went into a tunnel.
There are no sign-in screens anywhere, and most games don’t even request permission to send push notifications. Your progress will even automatically sync to the cloud so you can continue playing on your iPad, Mac, or Apple TV when you get home.
Oh, and you don’t need to worry about your privacy in all this either. Apple says that all of the games in the service have no third-party tracking software and don’t collect any advertising data. All that, to me, is certainly worth paying for.
The PD20 combines all of Astell & Kern’s longstanding understanding of electronics and metalwork with a freshly comprehensive suite of software that makes it a formidable option for headphone use on the move. It’s not cheap and it’s not small but the performance is good enough that many people won’t care
Superb headphone performance
Well-made and superbly finished
Flexible use options
Pricey
Large and heavy for portable use
Battery life in some modes is quite limited
Key Features
Advertisement
Review Price:
£1799
Storage
Advertisement
256gb internal storage expandable up to 2.25tb
DAC
Advertisement
Quad ESS DAC with twin headphone amplifier
Apps
Advertisement
Supports most major streaming services and Roon
Introduction
The market for personal audio players is a rather affair than it once was when the iPod dominated the business of listening to music on the move.
Astell & Kern has carved out a strong presence at the top of this more niche market and over the years, they have cleverly evolved their devices to be more than a portable device for use on the move.
Advertisement
The PD10 that I looked at in 2025 can deliver great performance over headphones or earphones but can then be parked in its supplied cradle and used as a superb source for your main system as well. It’s clever thinking and the PD10 is a superb product.
You might reasonably assume that a device called the PD20 would be more of the same; tweaked and improved but intended to offer the same dual use concept. The reality though is a bit different.
Advertisement
While Astell & Kern hasn’t abandoned the idea of being able to do more than use the PD20 out and about, it has revisited the functionality needed to make the PD20 a more effective partner for this primary role. As I shall cover, this includes functionality that hasn’t appeared on any Astell & Kern device up to this point as well.
Advertisement
As such, we need to see if this change in direction has yielded positive results and if the not inconsequentially expensive PD20 is the device you’ll want to turn to when it’s time to leave the house.
Price
The PD20 is available in the UK for £1,799. The dealer network isn’t huge but it covers most areas of the country and it is possible to purchased online if you are particularly remote. In the USA, it is available for $1,970 USD while in Australia, it will be $2,649.
Something that is important to stress is that the PD20 is not a replacement for the PD10. The two units will be sold together and this reflects the differences in specification and intended use between the two different models.
It’s also worth stressing that despite the higher number, the PD20 isn’t a higher specification device than the PD10 and indeed costs less.
Advertisement
Advertisement
Design
Superbly made and finished
A bit on the big and heavy side
Revised external controls
Clear and logical interface
Those of you familiar with Astell & Kern’s handiwork up to this point won’t find anything alarming about the PD20. The all metal chassis is solid, weighty (weighty enough that any of you who go in for a loose fitting garment might want to check what happens if you put it in a pocket before you head out) and finished to an immaculate standard.
There are some intriguing differences between the PD10 and PD20 though. The PD10 streamlines the controls to a row of buttons down the side. The PD20 by contrast has rather more going on in the form of both a pair of rotary controls on the top to control both tone and volume and a row of switches on the side; the functions of which we’ll cover in the specifications section.
Image Credit (Trusted Reviews)
There is no getting away from the fact that the Astell & Kern is pretty big for a ‘portable’ device. This might be better viewed as something that travels with you in a bag and, even then, it had better be a reasonable sized bag.
The upside to this is that it means that Astell & Kern is able to fit a decent sized screen which is bright and easy to read. It wasn’t too long ago that even the premium models didn’t feel as responsive and easy to use as a middling phone. This is very much no longer the case.
Advertisement
Advertisement
Image Credit (Trusted Reviews)
The interface itself is also as good as I’ve ever experienced from an Astell & Kern device. At its core is a version of Android but a great deal has been changed and tweaked by Astell & Kern to better suit its more specialised use case. Key things like swiping down to access key settings and menus has been retained though and it makes the PD20 feel wholly unintimidating to use.
Final mention has to go to the packaging too. It’s attractive, brilliantly laid out and does a fine job of making you feel like you haven’t wasted your money. You get some screen protectors included but no case which is a slight shame as I’m sure that your first scratch will be a mortifying experience.
Image Credit (Trusted Reviews)
Specifications
Limited internal memory with option to expand
Quad DAC arrangement
Adjustable amp classes
New hearing test software and EQ
Like the PD10, the PD20 offers 256GB of internal memory which, in 2026, feels a bit on the parsimonious side.
Advertisement
Yes, you can add a micro SD card of up to 2 terabytes in size and these have become more affordable in recent years but if you use Qobuz or Tidal as your main source of material (both of which the PD10 supports natively), these files can only be stored offline in the internal memory which will quickly fill up.
Image Credit (Trusted Reviews)
At the best part of two grand, 512gb doesn’t feel like too much of an ask. As well as Tidal and Qobuz other streaming services can be installed and the unit can also used via Roon and Roon ARC as well.
The decoding is significantly different to the PD10 which used AKM hardware. Here, there are no less than four ES9027PRO chips fitted that sum their results against each other, looking for and then correcting any decoding errors.
Advertisement
This is a lot of processing power (each DAC on its own can support eight channels of information) and the quoted performance is impressive. The DACs are externally clocked and fronted by an Octo Core processor.
The PD20 supports PCM to 768kHz and DSD512. There is also DAR (Digital Audio Remaster) software; an upsampling algorithm that can be used in PCM or DSD versions (and, no less importantly, can be switched off).
Advertisement
If you do run it, the sample rate increases considerably, with Astell&Kern stating that this allows for additional information to be interpolated and worked into the output. This is combined with six adjustable digital filters that adjust the roll off rate and amount of pre ringing.
Advertisement
Image Credit (Trusted Reviews)
The digital stage outputs to an internal amp that works a little differently to both the PD10 and indeed other Astell & Kern models. On the side of the casework you’ll find two three-position switches. The first allows you to switch the topology of the output between Class A, Class AB and Hybrid – intended to combine the benefits of both.
A second switch changes the amp current settings between high, medium and low, depending on the headphones you are using. This gives the Astell & Kern the notional ability to power some very demanding headphones indeed although if you run it in Class A mode and the high current setting, the battery life will be limited.
At the top of the chassis you’ll find both a 3.5mm and 4.4mm output for balanced and unbalanced headphones and these are coupled to a 150 step volume control which can be bypassed if you put the PD20 into line out mode. It also has an entirely effective two way Bluetooth implementation too. It’s still perfectly possible to use the Astell & Kern as a compact digital source for your system if the fancy takes you, even if the neatness of the PD10’s dock is absent.
Image Credit (Trusted Reviews)
Advertisement
So far, these are only minor changes from the PD10 but at this point, the PD20 starts to differ significantly. The PD20 is fitted with software that Astell & Kern has developed with Audiodo to perform an interactive listening test on your hearing and adjust the output of the PD20 based on your feedback.
Running the test takes less than five minutes and you can store multiple profiles on the device as well. In order for this result to be consistent, you’ll find a small pair of earphones in the box to run the test with that ensures the results will be consistent.
Advertisement
This is partnered both with tone control adjustment and an on board EQ to further tweak the performance to your needs. This is by far the most flexible and customisable Astell & Kern device I’ve tested and it should allow the PD20 to match up with pretty much any headphone and then adjust what it’s doing to suit your needs. Neither is it done there.
There is also software called Audiosphere which Astell&Kern claims ‘goes beyond conventional stereo, delivering a powerful sense of spatial immersion. It expands two-channel audio into a virtual three-dimensional sound field and offers four presets.’ I have to say it’s an acquired taste in use but it’s there if you want it.
Performance
Capable of excellent performance even with no EQ engaged…
… but it gets better still once you do
Switchable amp modes have less effect
Capable used as a line level source
Advertisement
Something that is very important to get across about the PD20 is that the new suite of software isn’t something that it ‘needs’ to have engaged to deliver a decent performance.
Leave it switched off, the EQ set to flat and none of the upsampling engaged and this is still a superb piece of audio equipment. Using it as a Roon Endpoint and sending it random corners of my music library did nothing to unsettle it and it’s as happy with the scuzzy shoegaze of New Meaning by Tempers as it is with the sublime high res of Sarah Jarosz’s Blue Heron Suite.
Advertisement
It’s tonally accurate and impressively good at keeping a feeling of space and three dimensionality in what you are hearing, even when you are using earphones.
Image Credit (Trusted Reviews)
If you do take the time out to run the hearing tests though, it does have a meaningful effect on the performance. For my sins, I am 45 and while I have looked after my hearing as best as I can, I am undoubtedly experiencing some rolling off at higher frequencies. After the tests had been run, there is a little more treble energy to the lovely Hallucinating Love by Maribou State that makes it a little more vibrant and exciting without it tipping over into being bright or harsh.
Something I found interesting is that I’ve lost a fraction more of the high end registers in one ear than the other and the test resulted in a gentle rebalancing of this which ensured the performance had a greater sense of stereo to it.
Advertisement
I was a little concerned that this would in turn affect adding crossfeed to the signal (as what was being transferred to the other channel would be boosted over what that ear ‘needs’ but I needn’t have worried and I still preferred the performance with some crossfeed engaged.
Advertisement
Image Credit (Trusted Reviews)
Against this, I have found the amp modes to be less profound in what they do. This is partly down to the reality that none of the headphones and earphones that live here are very hard to drive so the extra current on offer doesn’t have a night and day effect but, even allowing for this, I suspect that the digital tweaks will make more difference for most owners than the amp mode switches which I suspect will see less use once some initial tinkering has taken place.
There is also the good news that, while the PD20 doesn’t feel quite a slick to use in a fixed installation as the PD10, it’s still capable of delivering a truly satisfying performance.
Listening to Boards of Canada’s mighty Prophecy at 1420MHz via the PD20 captures the shifting and positively unsettling power of the work, the immediacy and scale it has and the intricate details of the samples are well worked into the main system.
The PD20 feels less a portable audio source than the PD10 but it doesn’t prevent it from being entirely effective used as one.
Advertisement
Advertisement
Should you buy it?
After a few years showing that their products could be more than a portable player, the PD20 feels like a revisiting of the concept. The extra software fitted means it is more flexible and capable of delivering a sparkling performance with pretty much any headphone or earphone at home or on the move.
This is still a big, heavy and expensive way of listening to music on the move and, as we head into summer with fewer pockets to hand, many people will find the PD20 simply too big and pricey to relax into using it out and about.
What it is might be best seen as a considered take on what a high spec portable player needs to be in 2026.
The overlapping nature of the Astell & Kern range means that this here to be a viable option for headphone only customers who might have been feeling left out by the system integration efforts of the PD10.
Advertisement
How We Test
We test every portable music player we review thoroughly over an extended period of time. We use industry standard tests to compare features properly. We’ll always tell you what we find. We never, ever, accept money to review a product.
Find out more about how we test in our ethics policy.
You must be logged in to post a comment Login