Impact of RMQA by IonQ on software side
It's been a while since I've been excited: IonQ has announced a new hardware technology, Reconfigurable Multicore Quantum Architecture (RMQA).
Although this technology has just been announced, I would like to consider the impact of this hardware technology on software. I've only seen the release of RMQA, but I'd like to look at it from a software point of view to see where it might be useful.
1、Simple increase in the number of qubits
2、It can be used for structured quantum circuits such as machine learning.
3、Even with a large number of qubits, it is relatively easy to create quantum entanglement over long distances.
Trapped Ion uses ionized atoms as qubits for calculations. Since there is less manufacturing error in the ions used, the variation of qubits is very small, and calculations can be performed with high accuracy. In this RMQA, a chain of 16 qubits (maybe some are used for cooling ions) is prepared as a lump, which is connected to each other to form a lump of 32 qubits, and then the calculation is performed with a large number of connections between the 32 qubits, and then back to the chain of 16 qubits. The amazing thing here is that you can create a chain of 32 qubits.
The next generation of IonQ machines will be 32 qubits fully coupled, and the quantum volume is expected to exceed an astonishing 4 million. This is far beyond the quantum volumes of existing machines.
The full coupling of 32 qubits is tremendous , and it is practically impossible to build a quantum computer with more connections than this. To perform similar calculations with superconducting or silicon qubits, we would need to make full use of two-qubit gates, create swap gates, and swap the states of the qubits to compensate for the number of couplings, but the current cost of two-qubit gates is so high that we are still a long way from being able to actually perform calculations. However, the current cost of two-qubit gates is too high, and it will be a long time before we can actually do the calculations. 32 qubits fully coupled as of 2021 will be a tremendous performance.
In 2018, IonQ has announced the creation of 79 qubit calculations and make stored 160 qubit ions. The idea of unit cells, where qubits have been fixed on a plane, has been introduced to ions.
This time, there are four chains of 16 qubits, but by increasing the number of chains of 16 qubits, we can prepare many qubits on a single chip while selecting qubits for calculation, combining them into 32 qubits for calculation, and then separating them again for calculation with another group. Simply by looking at how many groups of 16 qubits can be created in the future, the total number of qubits can be greatly increased. I have a feeling that we will be able to create more than 100 qubits (personal opinion).
Next is software, where full or multi-coupling of 32 qubits (24qubits) will come into play. The number of connections is important for the applications we are working on, such as QAOA used for optimization calculations, and especially for quantum machine learning. In addition, the order in which quantum circuits are calculated is important for the tensor network structure that we are using. RMQA and tensor networks seem to be a good match. The main structures used in tensor networks are MPS and MERA. MPS considers quantum entanglement by the coupling dimension with neighboring qubits, while quantum circuits can have a coupling dimension of $2^n$ for n qubits between neighboring virtual qubits. The same is true for MERA, but let's not get into the details, and simply say that by using RMQA, the advantages of tensor networks can be further enhanced.
I think this is a very useful technology for quantum machine learning, because by combining quantum circuits with quantum machine learning circuits, which are compatible with quantum circuits, we can create more efficient quantum circuits with more qubits.
It is important to note that while it is possible to create long-distance quantum entanglement while propagating adjacent to each other, by using the full coupling of 32 qubits, it is also possible to create quantum entanglement for long-distance qubits with a relatively small number of operations. If there are four chains of 16 qubits, quantum entanglement between the ends of 64 qubits can be created with a fairly small number of operations.
This is a new technology that could fundamentally change the way we think about quantum computer hardware, so I'm very excited about it, and I think it will make quantum machine learning even more exciting. That's all.