EE major here. All the equations in the third panel are classical electrodynamics. To explain the semiconductors needed to make the switches to make the, you really need quantum mechanics. You can get away with “fudged” classical mechanics for approximate calculations, but diodes and transistors are bona fide quantum mechanical devices.
Yes. You need specifically the condensed matter models, that have the added bonus of the equations making little sense without some interesting lattice designs at their side.
Semiconductors specifically have the extra nice feature of depending on time-dependent values, that everybody just hand-waves away and use time-independent ones because it’s too much magic.
Quantum Physics Postdoc here. Although technically correct this is also somewhat misleading. You need the band structure of solids, which is due to quantization and Pauli exclusion principle. The same quantum mechanics that explains why we did those strange electron energy levels for atoms in highschool. The majority of quantum mechanics, however, is not required: coherence, spin, entanglement, superposition. In the field we describe semiconductors as quantum 1.0, and devices that use entanglement and superposition (i.e. a quantum computer) as quantum 2.0, and smear everything else in-between. This
Good question. It would be application specific. I think evanescencnt wave coupling in EM radiation is considered " very classical" (whatever that actually means). But utilizing wave particle duality for tunneling devices is past quantum 1.0 (1.5 maybe?). However, superconductivity tunneling in Josephson junctions in a SQUID is closer to quantum 1.0, but 2.0 if used to generate entangled states for superconducting qbits for quantum computing.
Photonic processing, whilst very cool and super exciting, is not a quantum thing… Maxwells equations are exceedingly classical.
As for the rest it’s transistor design optimisation, enabled predominantly by materials science and ASMLs EUV tech I guess:), but still exploits the same underlying ‘quantum 1.0’ physics.
Spintronics (which could be what you mean by 2D) is for sure in-between (1.5?), leveraging spin for low energy compute.
Quantum 2.0 is systems exploiting entanglement and superposition - i.e. qubits in a QPU (and a few quantum sensing applications).
Oh my sweet summer child, a 100x yes, if only it were possible.
But more seriously, if you’re doing EE, the world of quantum is your oyster. Specialize in RF/MW design and implementation, we use it for qubit control, and you’ll be highly valuable.
Classical computers compute using 0s and 1s which refer to something physical like voltage levels of 0v or 3.3v respectively. Quantum computers also compute using 0s and 1s that also refers to something physical, like the spin of an electron which can only be up or down. Although these qubits differ because with a classical bit, there is just one thing to “look at” (called “observables”) if you want to know its value. If I want to know the voltage level is 0 or 1 I can just take out my multimeter and check. There is just one single observable.
With a qubit, there are actually three observables: σx, σy, and σz. You can think of a qubit like a sphere where you can measure it along its x, y, or z axis. These often correspond in real life to real rotations, for example, you can measure electron spin using something called Stern-Gerlach apparatus and you can measure a different axis by physically rotating the whole apparatus.
How can a single 0 or 1 be associated with three different observables? Well, the qubit can only have a single 0 or 1 at a time, so, let’s say, you measure its value on the z-axis, so you measure σz, and you get 0 or 1, then the qubit ceases to have values for σx or σy. They just don’t exist anymore. If you then go measure, let’s say, σx, then you will get something entirely random, and then the value for σz will cease to exist. So it can only hold one bit of information at a time, but measuring it on a different axis will “interfere” with that information.
It’s thus not possible to actually know the values for all the different observables because only one exists at a time, but you can also use them in logic gates where one depends on an axis with no value. For example, if you measure a qubit on the σz axis, you can then pass it through a logic gate where it will flip a second qubit or not flip it because on whether or not σx is 0 or 1. Of course, if you measured σz, then σx has no value, so you can’t say whether or not it will flip the other qubit, but you can say that they would be correlated with one another (if σx is 0 then it will not flip it, if it is 1 then it will, and thus they are related to one another). This is basically what entanglement is.
Because you cannot know the outcome when you have certain interactions like this, you can only model the system probabilistically based on the information you do know, and because measuring qubits on one axis erases its value on all others, then some information you know about the system can interfere with (cancel out) other information you know about it. Waves also can interfere with each other, and so oddly enough, it turns out you can model how your predictions of the system evolve over the computation using a wave function which then can be used to derive a probability distribution of the results.
What is even more interesting is that if you have a system like this where you have to model it using a wave function, it turns out it can in principle execute certain algorithms exponentially faster than classical computers. So they are definitely nowhere near the same as classical computers. Their complexity scales up exponentially when trying to simulate quantum computers on a classical computer. Every additional qubit doubles the complexity, and thus it becomes really difficult to even simulate small numbers of qubits. I built my own simulator in C and it uses 45 gigabytes of RAM to simulate just 16. I think the world record is literally only like 56.
Not at all. In a classical computer, the memory is based on bits of information which can either[1] in state 0 or state 1, and (assuming everything has been designed correctly) won’t exist in any other state than the allowed ones. Most classical computers work with binary.
In quantum computing, the memory is based on qubits, which are two-level quantum systems. A qubit can take any linear combination of quantum states |0⟩ and |1⟩. In quantum systems, linear combinations such as Ψ = α|0⟩ + β|1⟩ can use complex coefficients (α and β). Since α = |α|e^jθ is valid for any complex number, this indicates that quantum computing allows bits to have a phase with respect to each other. Geometrically, each bit of memory can “live” anywhere on the Bloch sphere, with |0⟩ at the “south pole” and |1⟩ at the “north pole”.
Quantum computing requires a whole new set of gates, and there’s issues with coherence that I frankly don’t 100% understand yet. And qubits are a whole lot harder to make than classical bits. But if we can find a way to make qubits available to everyone like classical bits are, then we’ll be able to get a lot more computing power.
The hardware works due to a quantum mechanical effect, but it is not “quantum” hardware because it doesn’t implement a two-bit quantum system.
[1] Classical computers can be designed with N-ary digit memory (for example, trinary can take states 0, 1, and 2), but binary is easier to design for.
then we’ll be able to get a lot more computing power.
I think that’s not quite true it depends on what you want to calculate. Some problems have more efficient algorithms for quantum computing (famously breaking RSA and other crypto algorithms). But something like a matrix multiplication probably won’t benefit.
It’s actually expected that matrix inversion will see a polynomial increase in speed, but with all the overhead of quantum computing, we only really get excited about exponential speedups such as in RSA decryption.
EE major here. All the equations in the third panel are classical electrodynamics. To explain the semiconductors needed to make the switches to make the, you really need quantum mechanics. You can get away with “fudged” classical mechanics for approximate calculations, but diodes and transistors are bona fide quantum mechanical devices.
But it’s also magic lol.
Yes. You need specifically the condensed matter models, that have the added bonus of the equations making little sense without some interesting lattice designs at their side.
Semiconductors specifically have the extra nice feature of depending on time-dependent values, that everybody just hand-waves away and use time-independent ones because it’s too much magic.
Quantum Physics Postdoc here. Although technically correct this is also somewhat misleading. You need the band structure of solids, which is due to quantization and Pauli exclusion principle. The same quantum mechanics that explains why we did those strange electron energy levels for atoms in highschool. The majority of quantum mechanics, however, is not required: coherence, spin, entanglement, superposition. In the field we describe semiconductors as quantum 1.0, and devices that use entanglement and superposition (i.e. a quantum computer) as quantum 2.0, and smear everything else in-between. This
Interesting. Does tunneling fall under 1.0 or 2.0? Isn’t it considered a property of classical electrical engineering?
Good question. It would be application specific. I think evanescencnt wave coupling in EM radiation is considered " very classical" (whatever that actually means). But utilizing wave particle duality for tunneling devices is past quantum 1.0 (1.5 maybe?). However, superconductivity tunneling in Josephson junctions in a SQUID is closer to quantum 1.0, but 2.0 if used to generate entangled states for superconducting qbits for quantum computing.
Clear as mud right?
It is now that I’ve looked up the different types of tunneling you mentioned. I didn’t know there were multiple types of tunneling before now.
Thanks for the informative reply and prompting me to do some reading!
You’re talking the old CPU designs, not the current ones fighting tunneling effects or the work-in-progress photonics & 2D hybrids.
No, I am not sure that I am.
Photonic processing, whilst very cool and super exciting, is not a quantum thing… Maxwells equations are exceedingly classical.
As for the rest it’s transistor design optimisation, enabled predominantly by materials science and ASMLs EUV tech I guess:), but still exploits the same underlying ‘quantum 1.0’ physics.
Spintronics (which could be what you mean by 2D) is for sure in-between (1.5?), leveraging spin for low energy compute.
Quantum 2.0 is systems exploiting entanglement and superposition - i.e. qubits in a QPU (and a few quantum sensing applications).
Can we trade?
Great write-up btw.
Oh my sweet summer child, a 100x yes, if only it were possible.
But more seriously, if you’re doing EE, the world of quantum is your oyster. Specialize in RF/MW design and implementation, we use it for qubit control, and you’ll be highly valuable.
This what?
Oh no! The [clever quantum mechanics joke] got him!
[radioactive decay triggered the poison gas?]
[Quantum hype train?]
[Imposter syndrome?]
Schrödinger opened the box 😱
Hmmm, so all computers are technically quantum computers?
Classical computers compute using 0s and 1s which refer to something physical like voltage levels of 0v or 3.3v respectively. Quantum computers also compute using 0s and 1s that also refers to something physical, like the spin of an electron which can only be up or down. Although these qubits differ because with a classical bit, there is just one thing to “look at” (called “observables”) if you want to know its value. If I want to know the voltage level is 0 or 1 I can just take out my multimeter and check. There is just one single observable.
With a qubit, there are actually three observables: σx, σy, and σz. You can think of a qubit like a sphere where you can measure it along its x, y, or z axis. These often correspond in real life to real rotations, for example, you can measure electron spin using something called Stern-Gerlach apparatus and you can measure a different axis by physically rotating the whole apparatus.
How can a single 0 or 1 be associated with three different observables? Well, the qubit can only have a single 0 or 1 at a time, so, let’s say, you measure its value on the z-axis, so you measure σz, and you get 0 or 1, then the qubit ceases to have values for σx or σy. They just don’t exist anymore. If you then go measure, let’s say, σx, then you will get something entirely random, and then the value for σz will cease to exist. So it can only hold one bit of information at a time, but measuring it on a different axis will “interfere” with that information.
It’s thus not possible to actually know the values for all the different observables because only one exists at a time, but you can also use them in logic gates where one depends on an axis with no value. For example, if you measure a qubit on the σz axis, you can then pass it through a logic gate where it will flip a second qubit or not flip it because on whether or not σx is 0 or 1. Of course, if you measured σz, then σx has no value, so you can’t say whether or not it will flip the other qubit, but you can say that they would be correlated with one another (if σx is 0 then it will not flip it, if it is 1 then it will, and thus they are related to one another). This is basically what entanglement is.
Because you cannot know the outcome when you have certain interactions like this, you can only model the system probabilistically based on the information you do know, and because measuring qubits on one axis erases its value on all others, then some information you know about the system can interfere with (cancel out) other information you know about it. Waves also can interfere with each other, and so oddly enough, it turns out you can model how your predictions of the system evolve over the computation using a wave function which then can be used to derive a probability distribution of the results.
What is even more interesting is that if you have a system like this where you have to model it using a wave function, it turns out it can in principle execute certain algorithms exponentially faster than classical computers. So they are definitely nowhere near the same as classical computers. Their complexity scales up exponentially when trying to simulate quantum computers on a classical computer. Every additional qubit doubles the complexity, and thus it becomes really difficult to even simulate small numbers of qubits. I built my own simulator in C and it uses 45 gigabytes of RAM to simulate just 16. I think the world record is literally only like 56.
Not at all. In a classical computer, the memory is based on bits of information which can either[1] in state 0 or state 1, and (assuming everything has been designed correctly) won’t exist in any other state than the allowed ones. Most classical computers work with binary.
In quantum computing, the memory is based on qubits, which are two-level quantum systems. A qubit can take any linear combination of quantum states |0⟩ and |1⟩. In quantum systems, linear combinations such as Ψ = α|0⟩ + β|1⟩ can use complex coefficients (α and β). Since α = |α|e^jθ is valid for any complex number, this indicates that quantum computing allows bits to have a phase with respect to each other. Geometrically, each bit of memory can “live” anywhere on the Bloch sphere, with |0⟩ at the “south pole” and |1⟩ at the “north pole”.
Quantum computing requires a whole new set of gates, and there’s issues with coherence that I frankly don’t 100% understand yet. And qubits are a whole lot harder to make than classical bits. But if we can find a way to make qubits available to everyone like classical bits are, then we’ll be able to get a lot more computing power.
The hardware works due to a quantum mechanical effect, but it is not “quantum” hardware because it doesn’t implement a two-bit quantum system.
[1] Classical computers can be designed with N-ary digit memory (for example, trinary can take states 0, 1, and 2), but binary is easier to design for.
I think that’s not quite true it depends on what you want to calculate. Some problems have more efficient algorithms for quantum computing (famously breaking RSA and other crypto algorithms). But something like a matrix multiplication probably won’t benefit.
It’s actually expected that matrix inversion will see a polynomial increase in speed, but with all the overhead of quantum computing, we only really get excited about exponential speedups such as in RSA decryption.