Every advanced chip today is patterned with EUV light — ultraviolet so intense it stops behaving like a smooth beam and starts behaving like individual raindrops.
That's the part almost nobody outside lithography knows, and it's the root cause of a defect problem the semiconductor industry has been stuck on for over a decade.
At the scale of a single transistor, the number of photons landing in one spot is down to random chance. Too few, and nothing forms — a piece of the circuit is just missing. Too many, or they scatter sideways, and two circuits that were supposed to stay separate fuse into one short.
Here's the part that makes it so stubborn: the obvious fix breaks itself. Turn up the light to eliminate the "missing" defects, and you get more bridging. Turn it down to stop the bridging, and missing defects come back. Every lab working on this and dozens of university groups are chasing a floor, not a fix. You can push the failure rate down. You can't push it to zero, because the cause isn't a machine error. It's quantum statistics.
The best current path isn't a single breakthrough — it's several partial ones stacked together: resist chemistry that absorbs light more efficiently so fewer photons are wasted, high-NA and hyper-NA optics that sharpen the image before the randomness has a chance to spread, and metal-oxide resists designed to react more sharply at the edge of a pattern instead of gradually. Each shaves the failure rate down. None of them zeroes it out.
I built a small animation to make this visible instead of abstract — each square is one exposure site getting the same dose of light, and the outcome is still different every time. Good, missing, or bridged, decided at random, live.
This is one of the most persistent open problems in semiconductor physics, and it only gets harder as chipmakers push past 2nm.
A MOSFET and a BJT both do the same job in a circuit: they switch a larger current on and off. But the way each one is told to switch is completely different.
A MOSFET is voltage-controlled. Put a voltage on the gate, and an electric field pulls together a thin conducting channel right under it, between the source and drain. The gate sits on top of a thin insulating oxide layer, so once that channel forms, essentially no current flows into the gate itself — it's holding the door open with a field, not a flow.
A BJT works the opposite way. It's current-controlled. A small, steady current has to keep flowing into the base terminal the entire time the device is on. That small base current controls a much bigger current flowing from collector to emitter — a real amplification, not just a switch. Cut the base current, and the bigger current stops too.
Same outcome — a bigger current switched on and off — reached through two different physical mechanisms: one held open by a voltage, one held open by a current. That distinction is why MOSFETs dominate digital logic (nearly zero standing gate current means far less wasted power at rest) while BJTs still show up wherever raw current-driven switching speed or analog gain matters.
A robot arm that's rigid all the way to its fingertip transmits every drivetrain imperfection straight through — gear backlash, torque ripple, and any sudden contact all arrive at the tip undamped. On collision with something unexpected, force can spike sharply before the control loop has time to react.
The fix used in many collaborative and humanoid robots is a series-elastic actuator: a deliberately compliant spring placed between the motor/gearbox and the load. Two things follow from that one spring. First, its deflection is a direct, cheap force measurement — Hooke's law, F = k·x, no separate force sensor needed. Second, the spring absorbs shock mechanically, before any sensor or software even measures it, simply because it physically gives.
Same contact event, two drivetrains: one transmits the shock, the other reads and absorbs it.
The RC time constant explained with a leaky bucket.
You fill a bucket under a tap. At first, water rushes in fast. Then it slows. Not because the tap closes — but because the pressure pushing water in shrinks as the bucket fills.
That's exactly how a capacitor charges through a resistor.
The resistor restricts flow, the way a narrow tap controls fill rate. The capacitor stores charge, the way a bucket holds volume. Voltage across the capacitor is water height. Current through the circuit is flow rate — and it falls as the gap between supply voltage and capacitor voltage narrows.
The curve that describes all of this: V(t) = Vs · (1 − e^(−t/τ))
One number captures the whole story: τ = RC
At t = τ, the capacitor reaches 63.2% of final voltage. Not 50%, not 100% — 63.2%. That specific number comes directly from 1 − 1/e, and it's the fingerprint of every RC circuit you'll ever touch.
Timers, filters, signal shapers, sensor interfaces — they all live by this constant.
The math is one line. The intuition is a bucket and a tap.
──
🔬 Next time I see a smooth curve on an oscilloscope, I'd remember what τ is.
One Photon Breaks a Superconductor, on Purpose - SNSPD (Superconducting Nanowire Single-Photon Detector)
A superconducting nanowire single-photon detector runs a thin wire, cooled to a few kelvin, right at the edge of its own superconductivity. Bias it with a steady current, just under the threshold that would destroy the superconducting state on its own, and the wire carries that current with zero resistance, indefinitely.
Then a single photon lands on it. That's enough. The photon deposits just enough local energy to knock a small patch of the wire out of its superconducting state, and for a brief moment that patch behaves like an ordinary resistor. The steady bias current can't disappear, so it's forced through that now-resistive patch anyway, and Ohm's law does the rest: current through resistance means a voltage appears, right there, right then. A simple readout circuit turns that voltage blip into one clean, sharp, countable pulse. The hotspot cools in nanoseconds, superconductivity returns, and the wire is ready for the next photon.
Nothing here multiplies or cascades. The entire signal is the local destruction of superconductivity itself, forced to show up as a pulse in an otherwise flat line. That's also why these detectors are prized for quantum work: the pulse is fast, clean, and has very little added noise, which matters a lot when the whole point is counting individual photons for quantum key distribution or other quantum-optics experiments.
Worth contrasting with a 2D-material avalanche photodetector that device turns one photon into a multiplying chain of freed electrons. This one turns one photon into a single, localized break in a current that was already flowing. Two different physical mechanisms, same job: make one photon countable.
Cut the power, and one of these memory cells forgets everything
Two memory cells, the same power cut. One of them loses its stored bit the instant power drops. The other does not move at all.
The first is a conventional volatile cell: it holds its bit as a small stored charge, and that charge needs continuous power to stay in place. The moment power is removed, the charge drains away and the bit resets to a default state — the information is gone.
The second is a memristor cell. Its stored value is not a charge; it is a physical resistance state set by the width of a conductive filament bridging an insulating layer. That filament does not depend on power to stay the way it is, so removing power changes nothing about the stored value.
Same interruption, two outcomes: one memory type needs to be kept running to remember anything, the other does not need power at all to hold what it already knows.
A superconducting nanowire single-photon detector runs a thin wire, cooled to a few kelvin, right at the edge of its own superconductivity. Bias it with a steady current, just under the threshold that would destroy the superconducting state on its own, and the wire carries that current with zero resistance, indefinitely.
Then a single photon lands on it. That's enough. The photon deposits just enough local energy to knock a small patch of the wire out of its superconducting state, and for a brief moment that patch behaves like an ordinary resistor. The steady bias current can't disappear, so it's forced through that now-resistive patch anyway, and Ohm's law does the rest: current through resistance means a voltage appears, right there, right then. A simple readout circuit turns that voltage blip into one clean, sharp, countable pulse. The hotspot cools in nanoseconds, superconductivity returns, and the wire is ready for the next photon.
Nothing here multiplies or cascades. The entire signal is the local destruction of superconductivity itself, forced to show up as a pulse in an otherwise flat line. That's also why these detectors are prized for quantum work: the pulse is fast, clean, and has very little added noise, which matters a lot when the whole point is counting individual photons for quantum key distribution or other quantum-optics experiments.
Worth contrasting with a 2D-material avalanche photodetector (an earlier post here): that device turns one photon into a multiplying chain of freed electrons. This one turns one photon into a single, localized break in a current that was already flowing. Two different physical mechanisms, same job: make one photon countable.
Open-source data-labeling tool — the unglamorous but essential first step of any AI project: tagging images, text, or audio to train custom models.
👥 *Who benefits:* 🔬 *Researcher* – Labeling defect images or datasets to train a custom detection model 👨💻 *Developer / Data Scientist* – Preparing high-quality training data without a paid labeling service 🎓 *Student* – Learning the real workflow behind building any machine-learning model 🏢 *Small Team* – Running a self-hosted labeling workflow with no per-seat licensing
Bharat S.
Every advanced chip today is patterned with EUV light — ultraviolet so intense it stops behaving like a smooth beam and starts behaving like individual raindrops.
That's the part almost nobody outside lithography knows, and it's the root cause of a defect problem the semiconductor industry has been stuck on for over a decade.
At the scale of a single transistor, the number of photons landing in one spot is down to random chance. Too few, and nothing forms — a piece of the circuit is just missing. Too many, or they scatter sideways, and two circuits that were supposed to stay separate fuse into one short.
Here's the part that makes it so stubborn: the obvious fix breaks itself. Turn up the light to eliminate the "missing" defects, and you get more bridging. Turn it down to stop the bridging, and missing defects come back. Every lab working on this and dozens of university groups are chasing a floor, not a fix. You can push the failure rate down. You can't push it to zero, because the cause isn't a machine error. It's quantum statistics.
The best current path isn't a single breakthrough — it's several partial ones stacked together: resist chemistry that absorbs light more efficiently so fewer photons are wasted, high-NA and hyper-NA optics that sharpen the image before the randomness has a chance to spread, and metal-oxide resists designed to react more sharply at the edge of a pattern instead of gradually. Each shaves the failure rate down. None of them zeroes it out.
I built a small animation to make this visible instead of abstract — each square is one exposure site getting the same dose of light, and the outcome is still different every time. Good, missing, or bridged, decided at random, live.
This is one of the most persistent open problems in semiconductor physics, and it only gets harder as chipmakers push past 2nm.
(NA stands for Numerical Aperture)
#Semiconductors #EUVLithography #ChipManufacturing #Nanotechnology #SemiconductorPhysics #Photonics #DeepTech #Engineering
3 days ago | [YT] | 0
View 0 replies
Bharat S.
Two ways to open the same door
A MOSFET and a BJT both do the same job in a circuit: they switch a larger current on and off. But the way each one is told to switch is completely different.
A MOSFET is voltage-controlled. Put a voltage on the gate, and an electric field pulls together a thin conducting channel right under it, between the source and drain. The gate sits on top of a thin insulating oxide layer, so once that channel forms, essentially no current flows into the gate itself — it's holding the door open with a field, not a flow.
A BJT works the opposite way. It's current-controlled. A small, steady current has to keep flowing into the base terminal the entire time the device is on. That small base current controls a much bigger current flowing from collector to emitter — a real amplification, not just a switch. Cut the base current, and the bigger current stops too.
Same outcome — a bigger current switched on and off — reached through two different physical mechanisms: one held open by a voltage, one held open by a current. That distinction is why MOSFETs dominate digital logic (nearly zero standing gate current means far less wasted power at rest) while BJTs still show up wherever raw current-driven switching speed or analog gain matters.
#Nanoelectronics #Semiconductors #Electronics #Physics #ScienceCommunication #Engineering
4 days ago | [YT] | 0
View 0 replies
Bharat S.
Actuator backlash and the spring that fixes it
A robot arm that's rigid all the way to its fingertip transmits every drivetrain imperfection straight through — gear backlash, torque ripple, and any sudden contact all arrive at the tip undamped. On collision with something unexpected, force can spike sharply before the control loop has time to react.
The fix used in many collaborative and humanoid robots is a series-elastic actuator: a deliberately compliant spring placed between the motor/gearbox and the load. Two things follow from that one spring. First, its deflection is a direct, cheap force measurement — Hooke's law, F = k·x, no separate force sensor needed. Second, the spring absorbs shock mechanically, before any sensor or software even measures it, simply because it physically gives.
Same contact event, two drivetrains: one transmits the shock, the other reads and absorbs it.
#Robotics #Mechatronics #ControlSystems #Nanoelectronics #Engineering #ActuatorDesign #Automation #STEM #ScienceCommunication #AppliedPhysics
4 days ago | [YT] | 0
View 0 replies
Bharat S.
#Nanoelectronics #Semiconductors #MOSFET #Microelectronics #DevicePhysics #ElectronicsEngineering #EngineeringEducation #VLSI #STEM #DigitalElectronics #LogicGates #CMOS #FlipFlop #SequentialLogic
4 days ago (edited) | [YT] | 0
View 0 replies
Bharat S.
The RC time constant explained with a leaky bucket.
You fill a bucket under a tap. At first, water rushes in fast. Then it slows. Not because the tap closes — but because the pressure pushing water in shrinks as the bucket fills.
That's exactly how a capacitor charges through a resistor.
The resistor restricts flow, the way a narrow tap controls fill rate. The capacitor stores charge, the way a bucket holds volume. Voltage across the capacitor is water height. Current through the circuit is flow rate — and it falls as the gap between supply voltage and capacitor voltage narrows.
The curve that describes all of this: V(t) = Vs · (1 − e^(−t/τ))
One number captures the whole story: τ = RC
At t = τ, the capacitor reaches 63.2% of final voltage. Not 50%, not 100% — 63.2%. That specific number comes directly from 1 − 1/e, and it's the fingerprint of every RC circuit you'll ever touch.
Timers, filters, signal shapers, sensor interfaces — they all live by this constant.
The math is one line. The intuition is a bucket and a tap.
──
🔬 Next time I see a smooth curve on an oscilloscope, I'd remember what τ is.
#Electronics #NanoElectronics #CircuitTheory #EmbeddedSystems #Engineering #Physics #Maths #VLSI #ScienceVisualization
1 week ago (edited) | [YT] | 0
View 0 replies
Bharat S.
One Photon Breaks a Superconductor, on Purpose - SNSPD (Superconducting Nanowire Single-Photon Detector)
A superconducting nanowire single-photon detector runs a thin wire, cooled to a few kelvin, right at the edge of its own superconductivity. Bias it with a steady current, just under the threshold that would destroy the superconducting state on its own, and the wire carries that current with zero resistance, indefinitely.
Then a single photon lands on it. That's enough. The photon deposits just enough local energy to knock a small patch of the wire out of its superconducting state, and for a brief moment that patch behaves like an ordinary resistor. The steady bias current can't disappear, so it's forced through that now-resistive patch anyway, and Ohm's law does the rest: current through resistance means a voltage appears, right there, right then. A simple readout circuit turns that voltage blip into one clean, sharp, countable pulse. The hotspot cools in nanoseconds, superconductivity returns, and the wire is ready for the next photon.
Nothing here multiplies or cascades. The entire signal is the local destruction of superconductivity itself, forced to show up as a pulse in an otherwise flat line. That's also why these detectors are prized for quantum work: the pulse is fast, clean, and has very little added noise, which matters a lot when the whole point is counting individual photons for quantum key distribution or other quantum-optics experiments.
Worth contrasting with a 2D-material avalanche photodetector that device turns one photon into a multiplying chain of freed electrons. This one turns one photon into a single, localized break in a current that was already flowing. Two different physical mechanisms, same job: make one photon countable.
#Nanoelectronics #QuantumTechnology #Superconductivity #Photonics #Physics #ScienceCommunication
1 week ago | [YT] | 0
View 0 replies
Bharat S.
#Semiconductor #NanoElectronics #DeviceCharacterization #Physics
1 week ago | [YT] | 0
View 0 replies
Bharat S.
Cut the power, and one of these memory cells forgets everything
Two memory cells, the same power cut. One of them loses its stored bit the instant power drops. The other does not move at all.
The first is a conventional volatile cell: it holds its bit as a small stored charge, and that charge needs continuous power to stay in place. The moment power is removed, the charge drains away and the bit resets to a default state — the information is gone.
The second is a memristor cell. Its stored value is not a charge; it is a physical resistance state set by the width of a conductive filament bridging an insulating layer. That filament does not depend on power to stay the way it is, so removing power changes nothing about the stored value.
Same interruption, two outcomes: one memory type needs to be kept running to remember anything, the other does not need power at all to hold what it already knows.
#Nanoelectronics #Semiconductors #Memristors #NonVolatileMemory
1 week ago (edited) | [YT] | 0
View 0 replies
Bharat S.
One photon breaks a superconductor, on purpose
A superconducting nanowire single-photon detector runs a thin wire, cooled to a few kelvin, right at the edge of its own superconductivity. Bias it with a steady current, just under the threshold that would destroy the superconducting state on its own, and the wire carries that current with zero resistance, indefinitely.
Then a single photon lands on it. That's enough. The photon deposits just enough local energy to knock a small patch of the wire out of its superconducting state, and for a brief moment that patch behaves like an ordinary resistor. The steady bias current can't disappear, so it's forced through that now-resistive patch anyway, and Ohm's law does the rest: current through resistance means a voltage appears, right there, right then. A simple readout circuit turns that voltage blip into one clean, sharp, countable pulse. The hotspot cools in nanoseconds, superconductivity returns, and the wire is ready for the next photon.
Nothing here multiplies or cascades. The entire signal is the local destruction of superconductivity itself, forced to show up as a pulse in an otherwise flat line. That's also why these detectors are prized for quantum work: the pulse is fast, clean, and has very little added noise, which matters a lot when the whole point is counting individual photons for quantum key distribution or other quantum-optics experiments.
Worth contrasting with a 2D-material avalanche photodetector (an earlier post here): that device turns one photon into a multiplying chain of freed electrons. This one turns one photon into a single, localized break in a current that was already flowing. Two different physical mechanisms, same job: make one photon countable.
#Nanoelectronics #QuantumTechnology #Superconductivity #Photonics #Physics #ScienceCommunication
1 week ago | [YT] | 0
View 0 replies
Bharat S.
4️⃣0️⃣
labelstud.io/
*LABEL STUDIO* 🏷️
Open-source data-labeling tool — the unglamorous but essential first step of any AI project: tagging images, text, or audio to train custom models.
👥 *Who benefits:*
🔬 *Researcher* – Labeling defect images or datasets to train a custom detection model
👨💻 *Developer / Data Scientist* – Preparing high-quality training data without a paid labeling service
🎓 *Student* – Learning the real workflow behind building any machine-learning model
🏢 *Small Team* – Running a self-hosted labeling workflow with no per-seat licensing
#AIToolADay #GrowWithAI
1 week ago | [YT] | 0
View 0 replies
Load more