Mémoire optique à changement de phase pour calcul neuromorphique destiné à l'intelligence artificielle
Bibliographic record
Abstract
RÉSUMÉ: L’objectif principal de ce projet de maîtrise est d’explorer l’usage de matériaux à changement de phase comme mémoire optique non-volatile dans le but d’effectuer des opérations matricielles rapidement et efficacement dans le domaine optique. Les multiplications sont effectuées par modulation d’amplitude grâce au contraste de pertes optiques entre la phase amorphe et cristalline du chalcogénure étudié, dont le changement de phase est rapide et réversible. L’état de ces mémoires, lorsque intégrées à un circuit photonique actif faisant également l’objet de ce mémoire, peut être individuellement altéré par l’emploi d’éléments chauffants intégrés de sorte à encoder les éléments de la matrice à multiplier. La proposition implique la caractérisation du Ge2Sb2Se5, un matériau émergent prometteur dans le domaine des mémoires non-volatiles optiques. Le contraste optique mesuré entre les états vitreux et cristallins du matériau implique un différence d’absorption approximative de 0.8 dB/µm dans la bande C et des pertes d’insertion négligeables dans l’état amorphe, justifiant son utilité potentielle comme mémoire. La méthode d’intégration proposée consiste au dépôt du matériau sur guides d’onde en silicium, de sorte à maximiser l’interaction entre le mode de propagation et le chalcogénure. Cette intégration a été effectuée par un procédé de microfabrication maison développé au Laboratoire de Microfabrication de Polytechnique Montréal. Un contraste optique de 0.5 à 1 dB/µm a été observé lorsque la mémoire est intégrée au circuit photonique suite à une cristallisation induite par recuit lent. Une nouvelle architecture de multiplication matricielle optique, mieux adaptée aux grandes matrices a été présentée et simulée, démontrant une amélioration de 3 ordres de grandeur en terme de pertes et d’imbalance de calcul pour des matrices de taile N = 100. ABSTRACT: The main objective of this Master’s project is to explore the use of phase-change materials as non-volatile optical memory to perform matrix operations rapidly and efficiently in the optical domain. Multiplications are achieved through amplitude modulation, taking advantage of the contrast in optical losses generated between the amorphous and crystalline phases of the studied chalcogenide material, which undergoes rapid and reversible phase changes. The states of these memories, when integrated into an active photonic architecture, which is also the subject of this project, can be individually altered using integrated micro-heaters, thereby encoding the elements of the matrix to be multiplied. The proposal involves the characterization of Ge2Sb2Se5, a promising and emerging material in the field of non-volatile optical memories. The measured optical contrast between the glassy and crystalline states of the material implies an approximate absorption difference of 0.8 dB/m in the C band with negligible insertion losses in the amorphous state, justifying its potential usefulness as memory. The proposed integration method involves depositing the material onto silicon waveguides to maximize the interaction between the propagation mode and the chalcogenide. This integration was carried out using a custom microfabrication process developed at the Microfabrication Laboratory of Polytechnique Montréal. An optical contrast of 0.5 to 1 dB/m was measured when the memory was integrated into the photonic circuit following crystallization induced by slow annealing. A new architecture for optical matrix multiplication, better suited for large matrices, was presented and simulated, demonstrating a three-orders-of-magnitude improvement in terms of computing losses and imbalance for matrices of size N = 100.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".