Comparing the largest quantum chips on the cloud in China and the US where the performance gap comes from

The ibm_sherbrooke system, a quantum computer in IBM's Quantum Cloud, now offers a new 127-quantum-bit Eagle processor that is optimized for error mitigation, the third version of the Eagle processor.

IBM says ibm_sherbrooke is its highest-performing system to date. The performance improvement is mainly reflected in the improvement of the coherence time. Compared to the ibm_washington system with the first generation of Eagle, the T1 coherence time went from 98.07 microseconds to 305 microseconds for sherbrooke, and the T2 time improved from 93.25 microseconds to 170.69 microseconds. The following figure.

Although IBM has implemented 433 quantum bits, the 127 quantum bit Eagle is still the largest quantum processor available to IBM customers. And what is the largest scale quantum processor available to China? It is not the 66-bit Zuchong II, nor is it the 121 bits of the Zhejiang University team reported by PhotonBox some time ago, because these processors are not yet available to the general public. According to PhotonBox research, the largest cloud-accessible quantum processor in China is the 50-qubit ScQ-P50 of the Quafu Quantum Computing Cloud Platform. as follows.

http://quafu.baqis.ac.cn/

The comparison reveals a large difference in the performance of the two quantum processors. In particular, the coherence time is 8.585 microseconds for the mean T1 and 10.471 microseconds for the mean T2 of the ScQ-P50 processor. The 18 quantum bit processor of the Quafu platform has a longer T1 coherence time with a mean of 35.492 microseconds. This is actually normal; as the number of quantum bits on a chip grows, crosstalk increases, resulting in shorter coherence times.

But IBM's Eagle r3 has surprisingly dramatic increases in coherence time as the number of quantum bits increases. For example, the ibm_ithaca system with 65 quantum bits of Hummingbird r3 has a median T1 coherence time of 191.01 microseconds, which is more than 100 microseconds less compared to Eagle r3, and a roughly comparable T2 time. The error rate is also lower for Eagle r3, with a two-quantum bit gate error rate of 0.006 compared to 0.009 for the 65-bit system.

How did IBM achieve these results?

IBM used a new calibration strategy. Previously, they designed the calibration procedure to minimize the error rate, but this came at the expense of stability. But starting with the sherbrooke system, they are changing this by emphasizing reduced measurement leakage, gate stability, and more consistent gate speeds. This will help eliminate drift and extend the time it takes to recalibrate again. Interestingly, this new calibration strategy may not necessarily provide the best quantum volume measurements, but is still preferable due to the increased stability.

With this change, IBM has implemented a different two-quantum-bit native gate, called the echoed cross-resonance (ECR) gate, to replace the previously used CNOT two-quantum-bit gate. the Qiskit translator has been updated to compile any existing Qiskit program to use the ECR gate. the topology of sherbrooke is shown below. Note that the double quantum bit ECR gate is only unidirectional (as shown by the connecting arrows below) and not bidirectional. Therefore, the transcoder will assign physical quantum bits when compiling the program.

The most efficient one-way mapping from quantum circuits to ibm_sherbrooke native gates.

For this upgrade, IBM concluded that the third revision of the Eagle processor uses a new architecture that provides higher quantum bit coherence. Despite the longer gate lengths, the increased coherence allows for lower gate errors than previously possible.