The title of this blog post is a quote of John Martinis, taken from the following talk he gave at Google LA on October of last year. The talk has only recently (Feb 28) been put on the web. Check it out. It’s about 1 hr long. I highly recommend it. It’s crystal clear and covers quite a lot of territory. If after you finish seeing it, you want even more detail, I recommend Martinis’ website at UCSB, where he has links to all the Doctoral theses of his past students. Doctoral theses are a treasure trove of information.
I predict that John Martinis’ QC designs will totally dominate quantum computer hardware R&D for the next 5-10 years.
D-Wave’s QC is fine and I wish them well. However, I specialize in writing QC software, and all the QC software I’ve ever written is for gate model QCs, not adiabatic QCs, so my heart is invested in the gate model. Perhaps history will prove that this is a good analogy:
D-Wave QC ~ analog computer,
Martinis’ QC ~ first vacuum tube computer like the Eniac.
There are other gate model QC designs out there besides Martinis’, but his design is already FAR more advanced than those of his competitors. Furthermore, his qubits are bigger (100 microns) and therefore easier to connect to with leads than those of competing QC designs (eg., ion trap). I never liked the fact that with the ion trap design, you have to physically move the qubits from location to location. Smells too mechanical to me. Mechanical means are always slower than electrical means. And they also break down more quickly.
Here are some brief notes I took from Martinis’ Google talk. I just wrote down some numerical values. I didn’t write down anything about all his wonderful explanations and diagrams of the physics and math that describes his QC.
What would M need to factor 2048 bit number?
2E8 physical qubits,
1 day runtime,
cost = $100-500 million,
length of machine = 1-10 meter
Fault Tolerant Error Correction With Surface Code
How efficient is surface code?
r_n = (number of physical qubits)/(number of logical qubits)
r_p = p/p_max = (probability of error per physical gate)/ (maximum allowed probability of error). p_max = 1E-2 for surface code. Other, less forgiving codes have p_max = 1E-3.
r_e = number of errors per unit of time
r_p= 1E-1, r_e= 1/age of universe ===> r_n=3000
Type of technology?
standard integrated circuit technology,
Al/Al-O/Al Josephson junctions on sapphire substrate
frequency = 5 GHz (= 240 mK, 2E-5 eV, 6E-2 meters)
temperature = 20 mK
His CNot quality?
40ns, 99.45% efficiency
Size of his current research group?
about 50 researchers, 1 Pomeranian dog
Why surface code?
“Surface code really looks quite ideal for building integrated circuits” because
(1) only 2D nearest neighbor interactions are required
(2) Has highest p_max, most forgiving, of all known error correction codes
Martinis’ 5 year plan?
Next 5 years, demonstrate 1000 physical qubit machine with control electronics at room temperature, outside of chip and cryostat.
Must bring control lines to qubits. Control lines separated by 100 microns. Have to bring 100-1000 control lines to edges of wafer.
Near 56:10 min mark, M says:”You have to think like a high energy physicist” (LHC detectors have a huge number of wires)
Martinis’ 10 year plan?
After achieve 1000 physical qubit machine, M plans to put control electronics right on the chip. Superconducting IC technology for doing this already exits, has existed for many decades. Recent advances in it made by D-Wave