Quantum computing has always struggled to achieve the necessary division of labor between processing and memory. While different approaches for separate RAM exist, they typically require bulky electromagnetic resonators. Scientists have now developed a computing architecture that stores memory as vibrations, exponentially shrinking the required hardware.
Researchers at ETH Zurich have created a high-overtone bulk acoustic-wave resonator that stores information as phonons, efficiently separating memory from processing. The architecture stores the working information as vibrations inside microscopic resonators while superconducting qubits handle processing as usual. The researchers say this could fit much more quantum information into a smaller space than electromagnetic memory, while keeping the information stable.
Standard computers, including everything from your phone to PCs and gaming consoles, separate memory and processing. The CPUs and GPUs perform calculations, while the RAM temporarily holds the required working data. Superconducting quantum computers typically don't implement this division of labor, with every qubit serving as both a processing element and a storage element. While superconducting qubits are excellent at processing, they have short memories and tend to “forget” the information they hold when the computer runs a long algorithm.
The current solution adds memory by wiring qubits to electromagnetic resonators, either as flat waveguide structures etched on the chip or three-dimensional microwave cavities. The problem is size. Superconducting quantum computers only work when chilled to a hair above absolute zero, requiring the whole machine to sit inside an ultra-cold refrigerator called a cryostat, where space is tight and cooling is expensive.
To store a qubit's state, the qubit transfers the information to a resonator, a container that traps a wave, where it's held as either zero or one microwave photon bouncing back and forth until the qubit takes it back. Now, the container has to be roughly as big as the wave it traps, and since a 5-GHz (quantum machines’ wavelength) microwave is centimeters long, every memory slot ends up as a centimeter-scale metal box crammed into that fridge. In summary, storing lots of quantum states can require bulky electromagnetic resonators.
The ETH scientist took a different approach. Instead of storing the quantum state as light, they used sound. Sound travels through solid material at a few thousand meters per second, while light travels at 300 million meters per second. At the same 5-GHz frequency, a sound wave's wavelength collapses from centimeters, as is the case with light, to a couple of micrometers, roughly four orders of magnitude smaller.
Based on this principle, the researchers built a high-overtone bulk acoustic-wave resonator (HBAR), a tiny acoustic device that uses mechanical vibrations, a.k.a. phonons, to hold quantum information. A transmon qubit is patterned onto one sapphire chip, with a second sapphire chip sitting above it. The chips are coupled via a small piezoelectric disc that flexes when the qubit's electric field changes.
This flex launches a sound wave straight down into the sapphire, where it bounces between the crystal's two flat faces, setting up a standing wave, much like a guitar string fixed at both ends. The crystal is only a few hundred micrometers thick, enabling the wave to make the round trip absurdly fast, and resulting in a dense ladder of harmonics spaced about 12.6 MHz apart. In the ETH device, each harmonic is a memory slot.
Thanks to this density, one resonator can hold a stack of information, each at a slightly different frequency, all inside a chip measuring 7.5 mm long, 2.5 mm wide, and 1 mm high, roughly the width of a small fingernail. Another advantage of the new architecture is its memory length. Measurements on the device put the transmon's coherence time at about 23 microseconds, while the phonon modes it talks to held their quantum states for between 127 and 368 microseconds. This means that the memory remembers roughly five to 16 times longer than the processor attached to it, which is precisely the division of labor you want.
When the processor needs a stored state, the qubit is tuned to resonance with the relevant vibration, and a swap operation transfers the state from the sapphire to the transmon. The team then either drives the transmon directly for a single-qubit operation, or detunes it slightly against a second vibration to accumulate a controlled phase shift, which is how they built two-qubit gates between modes that never touch each other directly. Finally, the modified state gets swapped back into its original mode to wait.
During testing, the single-mode gates achieved fidelities between 94.95% and 95.50%, with each swap operation costing about 1.71% on average. Their controlled-phase gates all cleared 85%, with the workhorse of the set landing at 89.2% after stripping out measurement errors. With those pieces in hand, they ran a three-mode quantum Fourier transform and then used it to execute a period-finding algorithm, correctly recovering the period of the test functions they fed it.
The researchers chose these algorithms deliberately, describing the quantum Fourier transform as “a fundamental computational procedure required for many quantum algorithms,” while period finding provided a more involved demonstration of the architecture combining quantum processing and memory.
The current demonstration proves the viability of the researchers' acoustic quantum hardware. However, the architecture is still limited by the number of phonon modes that can interact effectively with the transmon. In addition, scaling the architecture while maintaining precise control and low error rates remains a major challenge.
The team is now working to improve coherence, speed up transmon readout, and investigate alternative versions of the hybrid architecture. If those efforts pan out, the result could be hardware-efficient quantum random-access memory that eliminates the need for bulky electromagnetic components. The study is published in the journal Science.
Source: ETH Zurich
