RESEARCH SPOTLIGHT

“This is one of my favorite results so far! Because we reveal a gap of ‘ignorance' in quantum learning models, as it impacts the interpretability of a learning model. In this way, quantum learning models exhibit a fundamental interpretability limit.

See, interpretability of a learning model, be it classical or quantum, is the degree of model output understanding at a human level. And yes, we can mathematically model human understanding. The outputs of learning models are not always intuitive, correct, or even interpretable. In fact, there is a correlation between how interpretable a model is versus the accurate results it is able to produce.

Obviously, I wanted to understand quantum models along the same interpretability lines. Due to the probabilistic nature of quantum measurement, there isn’t a deterministic approach to classification, hence the ‘decision boundary' becomes a 'decision region.’ In this way, we quantise and simulate one of the widespread classical interpretability techniques in an autoencoder. Incidentally, the region mapped around a single data point, represents the regime in which data points are non-interpretable.

Interpretability of learning models is crucial to establishing social trust in AI-backed systems. It takes a technology that’s high risk, unproven, untested, and makes it well-behaved and manageable. For quantum learning models, this work marks only the beginning of our understanding of their intricate behavior. It’s exciting!"

"In our last publication, we established the potential of bilayer graphene (BLG)  as a host material for qubits. The qubits of choice are made from electrons confined in a double quantum dot that can be formed electrostatically in the material.

Apart from the more traditional way of encoding a qubit using the electron spin degree of freedom, we can use its  ‘valley’ state to encode quantum information. To ensure the efficacy of these qubits, it is imperative to understand the duration for which the confined electrons can maintain their spin and valley states, known as T1 times, which is ideally long. Graphene presents itself as an optimal candidate, due to minimized sources of noise associated with its microscopic details, which can damage the life-time of the qubits.

A significant breakthrough in our research is the successful demonstration of single-shot readout of both spin and valley Pauli spin blockade of a singlet-triplet double quantum dot in BLG. This allows us to perform the associated measurement of characteristic spin and valley relaxation times T1.
Remarkably, we find these durations to be comparable to those of state-of-the-art benchmarks, affirming the viability of graphene as a platform for quantum computing."

"In this paper, we tackle a challenge inherent to the platform of implanted donor spin qubits in silicon, which is to precisely place the donors near the silicon surface for effective qubit control, coupling and readout.

During implantation, we detect the arrival of a donor in the silicon substrate using on-chip detectors. However, the confidence of this detection is compromised at low implantation energies. Striking a balance is crucial, as low energies are essential to ensure donors are implanted near the surface with a small uncertainty.

Our team has developed a novel method to implant PF2 molecule ions to boost the P donor placement precision without compromising the detection signal. With molecule ions, you can have your cake and eat it: low energy of the donor ion you care about for precise, near-surface qubit placement and a large impact signal for reliable single ion detection. This work overcomes a hurdle on the road to building a full-scale quantum computer using implanted donor spin qubits in silicon."

"In our latest publication, we demonstrate coherent control of four interacting singlet-triplet qubits in a 2D germanium quantum dot array; the most extensive fully electrically controlled system of semiconductor spin qubits to date.

A total of four qubits can be individually initialized, manipulated, and readout. Transfer of classical and entangled states is achieved across the 2D-array. In achieving this, we overcome key challenges faced by quantum semiconductor processors because our qubits are operated using electric low frequency (base-band) pulses, which avoid error channels characteristic of microwave-based devices such as leakage and crosstalk.

With four universally controlled qubits in a bilinear array, these results put baseband-controlled singlet-triplet spin qubits in germanium firmly on the map as a potential candidate for large- scale quantum computing. Moreover, the device's 2D topology, together with individual qubit control, offers a unique opportunity to explore intriguing quantum many-body phenomena, including quantum phase transitions."

‘’In the rapidly growing field of semiconductor spin qubits, opportunities are still arising for new research avenues. So far, semiconductor qubits are typically created by confining electrons or holes in planar two-dimensional quantum dot arrays. Inspired by research done on bilayer germanium heterostructures, I asked the question: could we leverage the extra dimension to confine single holes in parallel planes? This question has a few motivations. Firstly, forming three-dimensional lattices of quantum dots to host qubits offers compact scaling opportunities and higher connectivity. Second, novel physics may emerge from the unique coupling configurations; for example, one could envision making electron-hole bilayers to simulate exciton condensation.

Typically in planar devices, we are able to confine single holes (or electrons) using electrostatic gating, where electric fields from a so-called plunger gate create a quantum dot within which holes are trapped. In this paper, we demonstrate for the first time that it is possible to use a single plunger gate to form vertically coupled double quantum dots (DQD), on bilayer germanium heterostructures. To find the position of the quantum dots, we perform triangulation by analyzing the relative electrostatic coupling of the dots to the surrounding gates and confirm that the DQD is formed in the z direction.

Currently, we are extending this method to form a 3D array (x-y-z) by accumulating double quantum dots beneath four plunger gates. While the research opportunities arising by the addition of another layer are vast, so are the challenges.
Gaining good control to form qubits in 3D lattices will require smart gate configurations and heterostructure design’’.