This technology introduces an advanced scheduling strategy for optimizing multi-model AI workloads on heterogeneous chiplet-based multi-chip modules (MCMs), aiming at maximizing performance efficiency.

Researchers at the University of California, Davis, have developed a memory system that uses optical interconnects.

      Biologics are antibodies produced by genetically engineered cells and are widely used in therapeutic applications. Examples include pembrolizumab (Keytruda) and atezolizumab (Tecentriq), both employed in cancer immunotherapy as checkpoint inhibitors to restore T- cell immune responses against tumor cells. These biologics are produced by engineered cells in bioreactors in a process that is highly sensitive to the bioreactor environment, making it essential to integrate process analytical technologies (PAT) for closed-loop, real-time adjustments. Recent trends have focused on leveraging integrated circuit (IC) solutions for system miniaturization and enhanced functionality, for example enabling a single IC that monitors O2, pH, oxidation-reduction potential (ORP), temperature, and glucose levels. However, no current technology can directly and continuously quantify the concentration and quality of the produced biologics in real-time within the bioreactor. Such critical measurements still rely on off-line methods such as immunoassays and mass spectrometry, which are time-consuming and not suitable for real- time process control.       UC Berkeley researchers have developed a microsystem for real-time, in-vivo monitoring of antibody therapeutics using structure-switching aptamers by employing an integrator-based readout front-end. This approach effectively addresses the challenge of a 100× reduction in signal levels compared to the measurement of small-molecule drugs in prior works. The microsystem is also uniquely suited to the emerging paradigm of “living pharmacies.” In living pharmacies, drug-producing cells will be hosted on implantable devices, and real-time monitoring of drug production/diffusion rates based on an individual’s pharmokinetics will be crucial.

      Current clinical practice for detecting low-concentration molecular biomarkers requires sending samples to centralized labs, leading to high costs and delays. Successful point-of-care (POC) diagnostic technology exist, such as the paper-based lateral-flow assay (LFA) used for pregnancy tests and SARS-CoV-2 rapid antigen tests, or miniaturized instruments such as the Abbot i-Stat Alinity. However, the former provides binary results or limited quantitative accuracy, and the latter is too expensive for in-home deployment. A promising approach for POC diagnostics, offering tailored circuit optimization, multiplexed detection, and significant cost and size reductions, is millimeter-sized CMOS integrated circuits coupled with microfluidics. Recent demonstrations include protein, DNA/RNA, and cell detection. The current complexity of system packaging (e.g., wire/flip-chip bonding) makes integrating microfluidics with more sophisticated functions challenging, and often-required syringe pumps and tubing are operationally unfriendly, limiting current approaches.       UC Berkeley researchers have developed a fully integrated, multi-mode POC device that requires single-step assembly and operates autonomously. Drawing inspiration from RFID technology and implantables, they have introduced inductively-coupled wireless powering and communication functionality into a CMOS bio-analyzer. With the chip being fully wireless, the die can be easily integrated into a substrate carrier, achieving a completely flat surface that allows for seamless bonding with the microfluidic module. In the final product, the device will be sealed in a pouch inside a vacuum desiccator. The user tears the pouch, adds a drop of sample, and the system automatically begins operation. The operation window can last up to 40 minutes, making the process insensitive to time delays. The present CMOS bio-analyzer integrates pH-sensing and amperometric readout circuits for both proton-based and redox-based immunoassays.

      Integrating microelectronics with microfluidics, especially those implemented in silicon-based CMOS technology, has driven the next generation of in vitro diagnostics. CMOS/microfluidics platforms offer (1) close interfaces between electronics and biological samples, and (2) tight integration of readout circuits with multi-channel microfluidics, both of which are crucial factors in achieving enhanced sensitivity and detection throughput. Conventionally bulky benchtop instruments are now being transformed into millimeter-sized form factors at low cost, making the deployment for Point-of-Care (PoC) applications feasible. However, conventional CMOS/microfluidics integration suffers from significant misalignment between the microfluidics and the sensing transducers on the chip, especially when the transducer sizes are reduced or the microfluidic channel width shrinks, due to limitations of current fabrication methods.       UC Berkeley researchers have developed a novel methodology for fabricating microfluidics platforms closely embedded within a silicon chip implemented in CMOS technology. The process utilizes a one-step approach to create fluidic channels directly within the CMOS technology and avoids the previously cited misalignment. Three types of structures are presented in a TSMC 180-nm CMOS chip: (1) passive microfluidics in the form of a micro-mixer and a 1:64 splitter, (2) fluidic channels with embedded ion-sensitive field-effect transistors (ISFETs) and Hall sensors, and (3) integrated on-chip impedance-sensing readout circuits including voltage drivers and a fully differential transimpedance amplifier (TIA). Sensors and transistors are functional pre- and post-etching with minimal changes in performance. Tight integration of fluidics and electronics is achieved, paving the way for future small-size, high-throughput lab-on-chip (LOC) devices.

      Molecular dynamics (MD) is a computational materials science modality widely used in academic and industrial settings for materials discovery and more. A critical aspect of modern MD calculations are machine learning interatomic potentials (MLIPs), which learn from reference quantum mechanical calculations and predict the energy and forces of atomic configurations quickly. MLIPs allow for more accurate and comprehensive exploration of material/molecular properties at-scale. However, state-of-the-art MLIP methods mostly use a short-range approximation, which may be sufficient for describing properties of homogeneous bulk systems but fail for liquid-vapor interfaces, dielectric response, dilute ionic solutions with Debye-Huckel screening, and interactions between gas phase molecules. Short-range MLIPs neglect all long-range interactions, such as Coulomb and dispersion interactions.      To address the current shortcoming, UC Berkeley researchers have developed a straightforward and efficient algorithm to account for long-range interactions in MLIPs. The algorithm can predict system properties including those with charged, polar or apolar molecular dimers, bulk water, and water-vapor interfaces. In these cases standard short-range MLIPs lead to unphysical predictions, even when utilizing message passing algorithms. The present method eliminates artifacts while only about doubling the computational cost. Furthermore, it can be incorporated into most existing MLIP architectures, including potentials based on local atomic environments such as HDNPP, Gaussian Approximation Potentials (GAP), Moment Tensor Potentials (MTPs), atomic cluster expansion (ACE), and MPNN (e.g., NequIP, MACE).

An innovative system designed to enhance manufacturing efficiency through predictive maintenance using machine learning.

      The demand for power electronics with smaller volumes, lighter weights, and lower cost has motivated ongoing investigation into alternative power passive component technologies. Miniaturization of power converters is bottlenecked by magnetics, whose power densities fundamentally reduce at small scales. Capacitors exhibit much more favorable densities at small sizes, but efficient voltage regulation and galvanic isolation are difficult to achieve without magnetics. Therefore piezoelectric components have emerged as compelling alternative passive components for power electronics. However,  their high-performance capabilities have been limited to applications of high load impedance due to the high characteristic of piezoelectric resonators (PRs) themselves.       To overcome this challenge, UC Berkeley researchers have developed novel piezoelectric resonator (PR) designs based on overtones, with enhanced power densities and reduced optimal load impedances. The overtone PRs have been demonstrated to have comparable efficiency to fundamental-mode PRs, while their capabilities for power handling density and lower optimal load impedances are increased. Use of overtone PRs can expand the utility of piezoelectrics to a wider scope of power electronics.

Brief description not available

Volatile organic compounds (VOCs) are released by various products and during various processes. Ethanol is one such VOC that is released as an important byproduct of alcoholic fermentation. Ethanol emitted during fermentation can be estimated using the amount of liquid lost during storage. The instrumentation needed to accurately quantify ethanol emissions is specialized and costly. Researchers at UC Santa Cruz have developed low-cost VOC sensors that are useful for the wine industry, among others.

 The dynamic patterning of 3D optical point clouds has emerged as a key enabling technology in volumetric processing across a number of applications. In the context of biological microscopy, 3D point cloud patterning is employed for non-invasive all-optical interfacing with cell ensembles. In augmented and virtual reality (AR/VR), near-eye display systems can incorporate virtual 3D point cloud-based objects into real-world scenes, and in the realm of material processing, point cloud patterning can be mobilized for 3D nanofabrication via multiphoton or ultraviolet lithography. Volumetric point cloud patterning with spatial light modulators (SLMs) is therefore widely employed across these and other fields. However, existing hologram computation methods, such as iterative, look-up table-based and deep learning approaches, remain exceedingly slow and/or burdensome. Many require hardware-intensive resources and sacrifices to volume quality.To address this problem, UC Berkeley researchers have developed a new, non-iterative point cloud holography algorithm that employs fast deterministic calculations. Compared against existing iterative approaches, the algorithm’s relative speed advantage increases with SLM format, reaching >100,000´ for formats as low as 512x512, and optimally mobilizes time multiplexing to increase targeting throughput. 

Certain difficult optimization problems, such as the traveling salesman problem, can be solved using so-called analog Ising machines, in which electronic components (such as certain arrangements of diodes or electronic switches) implement an analog of a well-studied physical system known as an Ising machine. The problem is recast so that its solution can be read off from the lowest-energy configuration of the analog Ising machine, a state which the system will naturally evolve towards. While promising, this methodology suffers major drawbacks. Firstly, the number of subunits, known as “spins”, in the analog Ising machines, as well as the number of connections between these subunits, can grow substantially with problem size. Secondly, existing implementations of this principle rely on chip constructions which are optimized for one or a few problems, and are not sufficiently reprogrammable to be repurposed efficiently for other applications. To address these problems, researchers at UC Berkeley have developed a device known as a Field-programmable Ising machine which can be adapted to implement an analog Ising machine using a variety of hardware designs, such as the diodes and switches mentioned above. These Ising machines can be effectively reprogrammed to efficiently solve a wide array of problems across various domains. The inventors have shown that this design can be applied to SAT (“Satisfiability”) problems, a class known to be similar to the traveling salesman problem, in that the number of spins needed and their level of connectivity do not grow too quickly with problem size.

Brief description not available

Modern Graphical Processing Units (GPUs) consist of several Streaming Multiprocessors (SM) – each has its own Register File (RF) and a number of integers, floating points and specialized computational cores. GPU program is decomposed into one or more cooperative thread arrays that are scheduled to the SMs. GPUs invest in large RFs to enable fine-grained and fast switching between executing groups of threads. This results in RFs being the most power hungry components of the GPU. The RF organization substantially affects the overall performance and energy efficiency of the GPU.

Brief description not available

The inventors have discovered a new way to generate ultrafast back-gating, by leveraging the surface band bending inherent to many semiconductor materials. This new architecture consists of a standard bulk semiconductor material and a layered material on the surface. Optical pulses generate picosecond time-varying electric fields on the surface material. The inventors have successfully applied this method to a quantum well Rashba system, as this is considered today one of the most promising candidates for spin-based devices, such as the Datta Das spin-transistor. The technology can induce an ultrafast gate and drive time-dependent Rashba and quantum well dynamics never observed before, with switching faster than 10GHz. This approach minimizes lithography and will enable light-driven electronic and spintronics devices such as transistors, spin-transistors, and photo-controlled Rashba circuitry. This method can be applied with minimal effort to any two-dimensional material, for both exfoliated and molecular beam epitaxy grown samples. Electric field gating is one of the most fundamental tuning knobs for all modern solid-state technology, and is the foundation for many solid-state devices such as transistors. Current methods for in-situ back-gated devices are difficult to fabricate, introduce unwanted contaminants, and are unsuited for picosecond time-resolved electric field studies.  

UCLA researchers in the Department of Electrical and Computer Engineering have developed a terahertz (THz) detector that utilizes the micro-Doppler effect to detect vibrations and long-distance sounds.

Electromagnetic noise from surfaces is one of the limiting factors for the performance of solid state and trapped ion quantum information processing architectures. This noise introduces gate errors and reduces the coherence time of the systems. Accordingly, there is great commercial interest in reducing the electromagnetic noise generated at the surface of these systems.Surface treatment using ion bombardment has shown to reduce electromagnetic surface noise by two orders of magnitude. In this procedure ions usually from noble gasses are accelerated towards the surface with energies of 300eV to 2keV. Until recently, commercial ion guns have been repurposed for surface cleaning. While these guns can supply the ion flux and energy required to prepare the surface with the desired quality, they are bulky and limit the laser access, making them incompatible with the requirements for ion trap quantum computing.To address this limitation, UC Berkeley researchers have developed an ion gun that enables in-situ surface treatment without sacrificing high optical access, enabling in situ use with a quantum information processor.

A new structure for III-N transistors that is able to maintain a high breakdown and operating voltage while improving the gain of the device.

Researchers led by Dr. Potkonjak from the UCLA Department of Computer Science have developed a technique to detect hardware Trojans in integrated circuits.

UCLA researchers in the Department of Materials Science & Engineering have developed a new method of diamond deposition in integrated circuit vias for thermal dissipation.

   Researchers at UCI have, for the first time, developed a method for modeling the efficiencies of artificial photosynthetic devices containing multiple light absorbers. As these devices more closely parallel naturally occurring photosynthesis, they offer higher performance than standard single-absorber devices.

A new approach to photonic integrated circuit fabrication.

UCLA researchers in the Department of Electrical Engineering have developed read-disturbance-free content addressable memory (CAM) using voltage controlled magneto-electric tunnel junctions (MEJs).

Researchers led by Jason Cong from the Computer Science Department at UCLA have developed a general synthesis and mapping system for SRAM-based FPGAs.