An analog-AI chip for energy-efficient speech recognition and transcription.

Models of artificial intelligence (AI) that have billions of parameters can achieve high accuracy across a range of tasks1,2, but they exacerbate the poor energy efficiency of conventional general-purpose processors, such as graphics processing units or central processing units. Analog in-memory computing (analog-AI)3-7 can provide better energy efficiency by performing matrix-vector multiplications in parallel on 'memory tiles'. However, analog-AI has yet to demonstrate software-equivalent (SWeq) accuracy on models that require many such tiles and efficient communication of neural-network activations between the tiles. Here we present an analog-AI chip that combines 35 million phase-change memory devices across 34 tiles, massively parallel inter-tile communication and analog, low-power peripheral circuitry that can achieve up to 12.4 tera-operations per second per watt (TOPS/W) chip-sustained performance. We demonstrate fully end-to-end SWeq accuracy for a small keyword-spotting network and near-SWeq accuracy on the much larger MLPerf8 recurrent neural-network transducer (RNNT), with more than 45 million weights mapped onto more than 140 million phase-change memory devices across five chips.

A standard test phantom for the performance assessment of magnetic resonance guided high intensity focused ultrasound (MRgHIFU) thermal therapy devices.

Purpose: Test objects for High Intensity Focused Ultrasound (HIFU) are required for the standardization and definition of treatment, Quality Assurance (QA), comparison of results between centers and calibration of devices. This study describes a HIFU test object which provides temperature measurement as a function of time, in a reference material compatible with Magnetic Resonance (MR) and ultrasound.Materials and methods: T-Type fine wire thermocouples were used as sensors and 5 correction methods for viscous heating artifacts were assessed. The phantom was tested in a MR-HIFU Philips Sonalleve device over a period of 12 months, demonstrating stability and validity to evaluate the performance of the device.Results: The study furnished useful information regarding the MR-HIFU sessions and highlighted potential limitations of the existing QA and monitoring methods. The importance of temperature monitoring along the whole acoustic path was demonstrated as MR Thermometry readings differed in the three MR plane views (coronal, sagittal, transverse), in particular when the focus was near a soft-tissue/bone interface, where there can be an MR signal loss with significant temperature and thermal dose underestimation (138% variation between the three plane views).Conclusions: The test object was easy to use and has potential as a valid tool for training, QA, research and development for MR guided HIFU and potentially ultrasound guided devices.

Publisher Correction: Memristive neural network for on-line learning and tracking with brain-inspired spike timing dependent plasticity.

A correction to this article has been published and is linked from the HTML and PDF versions of this paper. The error has been fixed in the paper.

Memristive neural network for on-line learning and tracking with brain-inspired spike timing dependent plasticity.

Brain-inspired computation can revolutionize information technology by introducing machines capable of recognizing patterns (images, speech, video) and interacting with the external world in a cognitive, humanlike way. Achieving this goal requires first to gain a detailed understanding of the brain operation, and second to identify a scalable microelectronic technology capable of reproducing some of the inherent functions of the human brain, such as the high synaptic connectivity (~104) and the peculiar time-dependent synaptic plasticity. Here we demonstrate unsupervised learning and tracking in a spiking neural network with memristive synapses, where synaptic weights are updated via brain-inspired spike timing dependent plasticity (STDP). The synaptic conductance is updated by the local time-dependent superposition of pre- and post-synaptic spikes within a hybrid one-transistor/one-resistor (1T1R) memristive synapse. Only 2 synaptic states, namely the low resistance state (LRS) and the high resistance state (HRS), are sufficient to learn and recognize patterns. Unsupervised learning of a static pattern and tracking of a dynamic pattern of up to 4 × 4 pixels are demonstrated, paving the way for intelligent hardware technology with up-scaled memristive neural networks.

Spike-timing dependent plasticity in a transistor-selected resistive switching memory.

In a neural network, neuron computation is achieved through the summation of input signals fed by synaptic connections. The synaptic activity (weight) is dictated by the synchronous firing of neurons, inducing potentiation/depression of the synaptic connection. This learning function can be supported by the resistive switching memory (RRAM), which changes its resistance depending on the amplitude, the pulse width and the bias polarity of the applied signal. This work shows a new synapse circuit comprising a MOS transistor as a selector and a RRAM as a variable resistance, displaying spike-timing dependent plasticity (STDP) similar to the one originally experienced in biological neural networks. We demonstrate long-term potentiation and long-term depression by simulations with an analytical model of resistive switching. Finally, the experimental demonstration of the new STDP scheme is presented.

An intravenous insulin protocol for strict glycemic control in acute ischaemic stroke.

BACKGROUND AND PURPOSE: There is a J-shaped association between admission glycemia and outcome. We designed an intravenous insulin protocol aiming at rapid and strict glucose control in hyperglycemic ischaemic stroke patients. Here, we describe the initial experience, safety, and efficacy of this protocol to achieve and maintain euglycemia in the first 48h. METHODS: The protocol is based on parallel scales for adjustment of insulin infusion rate according to current glycemia and the rate of change of glycemia, which was recommended in our stroke unit in 4/2007 in acute ischaemic stroke patients with glycemia >6mM. Data were registered in the Acute Stroke Registry and Analysis of Lausanne (ASTRAL). Capillary blood glycemia was measured hourly with fingerprick test at onset of treatment and after each scale change. Target glycemia was 4.0-6.0mM pre-prandially (5.5-8.0mM post-prandially). Hypokalemia was defined as serum potassium <3.5mM and measured every 12h. Specific algorithms were employed during meals and for patients leaving temporarily the stroke unit for diagnostic or therapeutic workup. RESULTS: In the 90 protocol patients, the first normoglycemia was achieved within 8h of treatment in 91.1% of patients (median interval 4h (interquartile range (IQR): 3-6). During the median treatment duration of 25.5h (IQR: 19.7-37.7), median glucose reduction was 2.5mM (IQR: 1.3-4.3mM). The overall rate of hypoglycemias was 4.5% and hypokalemias 18.5%. There was a significant increase in the proportion of hypokalemias on the first on-treatment measurement compared to admission (24.4% vs. 8.9%, P=0.002). CONCLUSIONS: The proposed intravenous insulin protocol controls acute post-stroke hyperglycemia but frequently leads to hypokalemia. This issue needs to be addressed for the protocol to be suitable for use in larger, randomized controlled trial to explore its clinical effect.

Primary ovarian rhabdomyosarcoma associated with clear cell carcinoma of the ovary: a case report and review of the literature.

The clinicopathologic and immunohistochemical findings of a case of coexistent primary ovarian rhabdomyosarcoma and clear cell carcinoma of the ovary are reported. The tumor was detected in a 41-year-old premenopausal woman who had a 1-year history of pelvic pain. On physical examination, a 14.0-cm left adnexal mass was found for which a hysterectomy with bilateral salpingo-oophorectomy and staging biopsies were performed. The tumor was composed of solid and cystic areas, and two distinct microscopic components were identified: clear cell carcinoma and rhabdomyosarcoma. Immunoperoxidase studies demonstrated that the neoplastic cells in the latter component were positive for desmin, muscle-specific actin, and myoglobin. A biopsy taken from the pelvic peritoneum revealed rhabdomyosarcoma (FIGO stage IIB). Despite two cycles of chemotherapy, the disease persisted in the pelvis 4 months after diagnosis.

Minocycline pigmentation of heart valves.

Minocycline, a derivative of tetracycline, is a broad spectrum antibiotic used in the treatment of gram-positive and gram-negative infections. Benitz et al. (1) were the first to report black discoloration of the thyroid gland in rats, dogs, and monkeys given minocycline. Since that time, there have been numerous reports in the literature describing minocycline related black pigmentation of the skin, thyroid gland, and other sites. We report an unusual case of minocycline induced pigmentation of the cardiac valves and coronary vessels. The pigment stained with Fontana-Masson and was reduced with bleaching. The exact nature of the pigment is unclear; however, various theories have been advocated. Ochronosis is another cause of black pigmentation of the heart valves; the clinical history should allow distinction between the two.