BT Research

Abstract Monolithic integration of active and passive photonics is the next step in miniaturization of semiconductor devices, to streamline their manufacturing and in doing so reduce their cost. However, Si’s inefficiency as a light emitter, and the high absorption from quantum wells or dots in III-V materials, requires that the two materials are heterogeneously integrated. This requires either regrowth or additional packaging steps, which add complication in scaling to mass manufacture, as well as high losses at material interfaces. This work proposes a solution whereby a monolithically grown GaAs/AlGaAs platform with both active and passive devices is achieved by using two vertically stacked waveguides, where only one of them contains quantum dots. In the gain section, where the InAs quantum dots emit light centred at 1300 nm, the waveguides are decoupled. By changing the relative environment of the two waveguides, light can be made to evanescently couple into the passive waveguide for low loss passive devices. This change can be brought about either by changing the local etch depth, or by the selective oxidation of Al0.98Ga0.02As underlayers. For the correct geometry, 94% of optical power can be successfully coupled between the active and passive waveguide. This coupling is simulated on FIMMPROP by Photon Design Ltd.

As society becomes increasingly connected, the demands placed on telecommunications systems will only grow. To meet these demands network providers want to deploy automated tools that make decisions based on available network information. Furthermore, there is a need for these tools to be agile, so that they can react to changes, or identify unexpected outcomes, as they occur in this rapidly evolving digital landscape. To address this challenge the first nonstationary contextual bandit method that simultaneously monitors the observed rewards for both changes and anomalies, SCAPA-UCB, is introduced. In addition to incorporating change and anomaly detection, the proposed approach relaxes common nonstationary bandit assumptions on the reward distribution for an arm, allowing contextual information to be incorporated using a broad range of statistical models. Furthermore, the method provides a faster retraining process once a change is detected. Extensive simulation studies are performed to establish the favourable performance of SCAPA-UCB, and an application categorising maintenance tasks on a telecommunications network is presented. • A novel nonstationary contextual bandit algorithm, SCAPA-UCB, is proposed. • The first contextual bandit method to detect changes and anomalies in rewards. • Relaxes common contextual reward assumptions such as linearity or Gaussian noise. • Simulations show excellent performance in a range of nonstationary settings. • Presents an application categorising telecoms network maintenance tasks.

Abstract Environmental assessments of digital services currently apply an accounting perspective, and for telecommunication networks (TN) allocate electrical energy consumption in proportion to data traffic. Yet, the power draw by wired TN infrastructure is almost independent of the volume of data traffic flowing through it. Previous assessments of the effect of data traffic on energy consumption thus tended to over‐estimate the short‐term impact on energy consumption.However, the growth of peak data traffic rates is a main driver of increasing TN bandwidth capacity and has an indirect impact on electrical energy consumption. This nuanced causal relationship has not been consistently represented in allocation approaches used for attributional carbon footprints.In this text, we apply a form of consequential system expansion by considering the long‐term response to peak‐traffic growth. This allows us to model long‐run marginal changes to product system attributes that are fixed in the short‐term. The outcome illustrates a causally consistent allocation approach that avoids contradicting the short‐term behavior of the engineered system.Based on a causal inference graph of the drivers for the fixed baseload power draw by TN, we distinguish between the effects of different types of data as they contribute to traffic peaks. From this, we develop transform functions that re‐allocate environmental burden to peak traffic. We present such functions for the specific case of periodically diurnal traffic in TN (including video‐on‐demand) and discuss the case of sporadic high‐throughput events (including video streaming of life sport events and games downloads).The allocation model incentivizes a reduction of peak demand through avoidance or demand‐shifting, to decelerate the long‐term expansion of TN infrastructure.

Abstract We report on the key design factors for the development of Type-II ‘W’-lasers for O-band (1260–1360 nm) applications. We investigate the effects of InGaAs and GaAsSb quantum well composition and thicknesses on the emission wavelength and recombination efficiency as well as of (Al, Ga) As barriers on optimum electrical and optical confinement. Photoluminescence (PL) tests structures and full device structures were fabricated and characterised. 1.25 µm emitting lasers were demonstrated with a threshold current density and J th values of 480 ± 10 A cm−2 at 290 K, whereas 1.3 µm lasers showed an increased J th value of 5.5–7 kA cm−2 at 290 K. The PL test structures exhibited a similar trend with decreasing intensity with increasing wavelength. Gain measurements of the 1.3 µm device demonstrate reasonably low optical losses of 10–15 c m − 1 and a threshold modal gain of ≈25 c m − 1 .

Telecoms operators are increasingly adopting principles of cloudification and virtualisation to optimise the delivery of existing services, while leveraging features of flexibility and elasticity to unlock value from new services. The extension of cloud infrastructure to deeper edge locations creates a challenge around optimal workload placement. One of the major issues to address is how to handle diverse and competing workload types – processor and memory hungry applications on the one hand, versus network-oriented applications on the other – within a resource-constrained edge compute footprint. While aiming to drive maximal performance, it is crucial to understand power consumption ramifications. This paper explores some fundamental performance and power tradeoffs linked to the placement of workloads across a dual socket edge server via CPU pinning in different ways. By devising a range of experimental test scenarios for distinct workload types, we reveal key insights. In one example, pinned placement of a process-intensive singleton workload using cores across sockets achieves 28% improved performance, but at the expense of 89W extra power consumption. In another example, pinned placement of workloads that are part of a service chain achieves 66% better performance and a power reduction of 11.4W when pinned to the same socket, rather than cross socket. Key insights such as these can inform either a rudimentary approach to workload placement based on heuristics or can provide the foundation for a more intelligence-based and automated approach.

This article introduces a server-centric cellular Passive Optical Network (C-PON) architecture to support the deployment of Augmented Reality (AR)/ Virtual Reality (VR) event viewing applications in edge data centers. The proposed architecture is compared with the state-of-the-art Spine-and-Leaf architecture. For fair comparison, we model production style environments based on both C-PON and Spine-and-leaf data center architectures. We developed a Mixed Integer Linear Programming (MILP) model with multi-objective function to optimize routing of AR/VR traffic on both C-PON and Spine-and-Leaf architectures. The multi-objective function considers minimizing power consumption and minimizing end-to-end delay within the network architectures. We compare hosting the AR/VR applications in C-PON and in Spine-and-Leaf in terms of the power consumption, the average delay in links, and the end-to-end delay per user. We also developed a heuristic algorithm to enhance the scalability enabling the optimization of complex and larger systems. The results show that C-PON can enable substantial savings in terms of power consumption compared to the state-of-the-art Spine-and-Leaf architecture.

A key challenge for telco cloud deployments is achieving a suitable balance of performance for the mix of workloads, and the power consumption of the compute infrastructure hosting those workloads. Recent generations of Intel x86 servers have a broad suite of technologies aimed at granular tuning of the multi-core processor resources: Cache Allocation Technology (CAT), Memory Bandwidth Allocation (MBA) and Speed Select Technology (SST) are key examples. Further fine-tuning is possible by exploiting frequency scaling properties of the "uncore" component of the CPU which include the last level cache and memory bandwidth control units- i.e., these are distinct from the part of the processor socket silicon footprint where "cores" reside. Such tuning capabilities can realise deterministic performance in the presence of mixed priority workloads which is a major consideration for more resource-constrained environments in the cloud compute continuum, e.g., the network edge where a mix of Containerised Network Functions (CNFs) can co-exist with other workload types. This paper explores a method of combining CAT, MBA, SST and uncore frequency tuning in tandem to achieve improved performance and power efficiency where a high priority workload co-exists with a "Noisy Neighbour" low priority workload. In one specific case, a performance increase of 29.6% is achieved in conjunction with a power reduction of 18.2% when compared with the baseline "untuned" set-up.

Abstract The numerical model has been developed for atomic ladder (cascade) systems, which proves to be suitable for Rydberg RF sensors.This model is designed with the capability to be extended to N-level non-thermal atomic systems with ease. The adaptability of this model to Rydberg RF sensors is validated by comparing its results with those of earlier reported experiments. The numerical model here provides a good approximation to the experimental results. Thus, such a dynamic model can be useful for quantum technologists in designing, optimising, and developing the Rydberg RF-based sensors in a user-friendly manner.

Abstract In the past years, energy consumption has increased rapidly due to many factors, including the rise in technology adoption. This has many downfalls, from higher costs to CO$$_2$$ 2 emissions. Human activities in offices and houses represent a considerable amount of energy usage. A digital twin (DT) of an open-plan common space is created, serving the purpose of remote room occupancy monitoring and automatic detection of energy consumption. A virtual reality (VR) model is developed and integrated to temperature, humidity and imaging sensors. For maintaining privacy, images are processed in local computers to measure occupancy levels and live video feed were never transmitted. The same set of imaging sensors were also used in a bespoke computer vision module for energy consumption estimation. The human avatars were mapped with high correlation (R$$^{2}$$ 2 $$= 0.85$$ = 0.85 ) with actual positions on floor. Our energy consumption algorithm accuracy obtained true positive rate of $$91.58\%$$ 91.58 % and F1 score of $$81.96\%$$ 81.96 % . Finally, all this information is transmitted and visualized to the 3D digital twin for remote monitoring and simulation.