Examensarbeten för kandidatexamen

Länka till denna samling:

https://odr.chalmers.se/handle/20.500.12380/51

Browse

Senast publicerade

Visar 1 - 5 av 25
  • Bild (thumbnail)
    Post
    QnAS - Quantum noisy Algorithm Simulator: Ett simuleringsverktyg för kvantalgoritmer med störningar i Python
    (2022) Blom, Axel; Edenmyr, Albin; Martinson, Edvin; Nordqvist, Ludvig; Palmqvist, Didrik; Wikman, Isak; Chalmers tekniska högskola / Institutionen för mikroteknologi och nanovetenskap (MC2); Lundgren, Per; Frisk Kockum, Anton; Debnath, Kamanasish
    Quantum computers are predicted to vastly outperform classical computers for certain calculations in the future and are therefore one of the most talked about research fields of our time. For the time being, quantum computers are however severly limited by the physical noises in the hardware. In this project a program called Quantum noisy Algorithm Simulator (QnAS) was developed in order to simulate quantum algorithms affected by relaxation, dephasing, excitation and unwanted interference between qubits. The aim was for the program to be able to simulate up to 15 qubits, be efficient and easy to use. The project was conducted on behalf of Wallenberg Centre for Quantum Technology and therefore had their specific implementation in mind, but the program could be generalised for other implementations of quantum computers as well. The program is written in Python and is based on the package QuTiP. The simulations are performed by the Monte Carlo wave function method in order to simulate as many qubits as possible. The result was a Python package that can be installed via pip. The program was verified by simulating a simple quantum algorithm and comparing the result with experimental data. Also presented in the report are results that demonstrate the functionality and performance of the program. The number of qubits that can be simulated is limited by the performance of the computer and it is considered reasonable to simulate up to 13 qubits on a personal computer whilst more than 15 qubits could be simulated on a computer cluster. The ambition is for the program to be used in the further development of quantum computers by for example giving information about noise resistance of algorithms, indicating which hardware parameters need to be improved and for troubleshooting experimental results. Some potential further developments that could increase the usability of QnAS are also discussed.
  • Bild (thumbnail)
    Post
    Maskininlärning för klassificering med en kvantdator
    (2022) Broback, Erik; Henkow, Victor; Liljenzin, Oscar; Lund, Andreas; Maltesson, Alex; Zaya, Emil; Chalmers tekniska högskola / Institutionen för mikroteknologi och nanovetenskap (MC2); Lundgren, Per; Fitzek, David; Frisk Kockum, Anton
    I detta arbete implementeras och undersöks en klassisk och två kvantmekaniska maskininlärningsalgoritmer för att se om kvantalgoritmerna har någon fördel gentemot den klassiska. De algoritmer som undersöks är en stödvektormaskin (SVM), en kvantmekanisk variationell klassificerare (VQC) och en kvantmekanisk kärnfunktionsuppskattare (QKE). De kvantmekaniska algoritmerna testas i huvudsak med få (⩽ 7) kvantbitar genom simuleringar i mjukvarubiblioteken PennyLane och Qiskit, men även till viss del på riktiga kvantdatorer från IBMQ. Efter implementering och testande med olika datamängder, kärnfunktioner och tillståndsförberedelser jämförs prestandan för de olika algoritmerna, genom bedömning av noggrannheten och körtiden för varje test. Resultaten visar både via noggrannhet och körtid att SVM:en presterar bäst, QKE:n presterar näst bäst och i ett fall bättre än SVM:en, och VQC:n presterar sämst. Slutsatsen är att denna jämförelse inte är till kvantalgoritmernas fördel, då den klassiska algoritmen redan löser klassificering väl. Förmodligen finns mer potential hos kvantalgoritmer när det kommer till att lösa andra problem som klassiska algoritmer inte kan hantera.
  • Bild (thumbnail)
    Post
    Optimering av tvåkvantbitsgrindar på en supraledande kvantdator
    (2021) Ingelsten, Emil; Nordqvist, Linus; Ozolins, Jesper; Reusch Eide, Alexander; Sahlberg, Elina; Zander, Henrik; Chalmers tekniska högskola / Institutionen för mikroteknologi och nanovetenskap (MC2); Vassilev, Vessen; Frisk Kockum, Anton; Fernández Pendás, Jorge
    Just nu pågår en kapplöpning om att bygga den största, snabbaste och mest stabila kvantdatorn. Förutom att realisera kvantbitarna själva på ett stabilt sätt är en annan stor utmaning att implementera kvantgrindarna som ska operera på kvantbitarna så perfekt som möjligt. Särskilt flerkvantbitsgrindar har visat sig vara svåra att implementera med tillfredsställande snabbhet och precision. I denna studie optimerade vi tvåkvantbitsgrindarna iSWAP ochCZ med avseende på grindfidelitet med hjälp av Python-paketet QuTiP. Mer specifikt gjordes detta genom att numeriskt modellera uppställningen som i nuläget används på Chalmers för att implementera tvåkvantbitsgrindar, bestående av två supraledande kvantbitar och en ställbar kopplare. Med denna uppställning appliceras tvåkvantbitsgrindar genom att modulera det externa magnetiska flödet (t) för att förändra den ställbara kopplarens karakteristiska frekvens. Vi lyckades uppnå en iSWAP-grind med fidelitet 0.9967 och operationstid 95 ns, och en CZ-grind med fidelitet 0.9991 och operationstid 114 ns. Dessa operationstider är avsevärt snabbare än nuvarande implementationer, och fideliteterna är betydligt högre än vad som tidigare realiserats.
  • Bild (thumbnail)
    Post
    Scalable signal mixer stages in quantum processors
    (2021) Andersson, Linus; Folkesson, Robin; Jonasson, Axel; Reiner, Sofia; Chalmers tekniska högskola / Institutionen för mikroteknologi och nanovetenskap (MC2); Vassilev, Vessen; Gasparinetti, Simone; Križan, Christian; Tancredi, Giovanna
    The development of cost-effective and compact quantum computers in the Quantum Technology Laboratory (QTL) at Chalmers University of Technology demands that the hardware of the control system can easily be scaled with the number of qubits. In this thesis the design and development of a rack instrument prototype is presented along with an evaluation of its performance. Three printed circuit boards were constructed and mounted in a subrack with the purpose of upconverting and downconverting the frequency of signals to and from the quantum processor with IQ mixers. Measurements performed at QTL confirmed image rejection and local oscillator signal suppression between 49-51 dB and 86-85 dB respectively for the calibrated mixers. The instrument was not found to add any measurable phase noise beyond what was already present in the signal source. Possible future improvements include integrating amplifiers to both up- and downconverting mixer stages. Including amplifiers will also allow for an additional local oscillator signal distribution card to be added, which would improve the phase stability in the quantum computing system.
  • Bild (thumbnail)
    Post
    Detektion av defekter i nanostrukturer med maskininlärning - En lösning med Autoencoders och Unsupervised Learning
    (2020) Carlsson, Eric; Cronquist, Olof; Eriksson, Oscar; Ulmestrand, Mattias; Chalmers tekniska högskola / Institutionen för mikroteknologi och nanovetenskap (MC2); Vassilev, Vessen; Lindvall, Niclas; Mahashabde, Sumedh
    This study aims to investigate the possibilities of using Machine Learning for detecting fabrication defects in nanostructures. The structures investigated are microwave circuits produced at the Department of microtechnology and nanoscience (MC2) at Chalmers University of Technology. Firstly, a particularly defective circuit was investigated with Logistic Regression, Dense Neural Networks, Convolutional Neural Networks as well a Transfer Learning based method with ResNetV2 and Principal Component Analysis. The distribution of defective and non-defective circuits was large and balanced enough to achieve 100 % correct classification of the sections with almost all models. Two more realistically defective circuits were further investigated, where the defective sections were widely underrepresented. However, a Convolutional Autoencoder (CAE), trained with either Supervised or Unsupervised Learning, was largely successful in separating defective sections from non-defective ones with a clear boundary based on the reconstruction error provided by the CAE. Furthermore, the CAE was in many cases able to locate the exact positions of defects by marking the areas of maximum reconstruction error, and also flag defective sections that previously was unsuccessful with manual inspection. The circuit sections used for the CAE were automatically sampled from larger images of circuits. After being sampled from the larger images of circuits, the sections were only downsampled and not preprocessed in any other way. The success of the unsupervised approach is the main achievement of the study, as all that is needed to train the model is completely uninvestigated images. The approach is estimated to save several days of manually inspecting whole circuits for defects.