Nxp B.v.

Hamed Fatemi

Jose De Jesus Pineda De Gyvez

Ajay Kapoor

Juan Diego Echeverri Escobar

Kristof Blutman

Jan Hoogerbrugge

Rinze Ida Mechtildis Peter Meijer

Jose Pineda De Gyvez

Sebastien Antonius Josephus Fabrie

Surendra Guntur

Ghiath Al-kadi

Ghiath Al-Kadi

Jurgen Geerlings

Luca Benini

Davide Rossi

Emad Ayman Taleb Ibrahim

Gonzalo Moro Pérez

Hendrik Corporaal

Manuele Rusci

Elisabetta Farella

Kristóf László Blutman

A human-machine-interface system comprising: register-file-memory, configured to store input-data; a first-processing-element-slice, a second-processing-element-slice, and a controller. Each of the processing-slices comprise: a register configured to store register-data; and a processing-element configured to apply an arithmetic and logic operation on the register-data in order to provide convolution-output-data. The controller is configured to: load input-data from the register-file-memory into the first-register as the first-register-data; and load: (i) input-data from the register-file-memory, or (ii) the first-register-data from the first-register, into the second-register as the second-register-data.

Human-machine-interface system comprising a convolutional neural network hardware accelerator

An apparatus for processing a signal for input to a neural network, the apparatus configured to: receive a signal comprising a plurality of samples of an analog signal over time; determine at least one frame comprising a group of consecutive samples of the signal, wherein the or each frame includes a first number of samples; for each frame, determine a set of correlation values comprising a second number of correlation values, the second number less than the first number, each correlation value of the set of correlation values based on an autocorrelation of the frame at a plurality of different time lags; provide an output based on the set of correlation values corresponding to the or each of the frames for a neural network for one or more of classification of the analog signal by the neural network and training the neural network based on a predetermined classification.

Apparatus for processing a signal

An electronic device including a power source providing a source voltage, a capacitor, a primary regulator circuit, an always-on load that is active during a low power mode, and a recycle control circuit. The primary regulator circuit receives the source voltage and has an output that maintains a charge on the capacitor during an active mode. The primary regulator circuit does not contribute to a charge on the capacitor during the low power mode. The recycle control circuit includes a select circuit and a select control circuit. The select circuit selects, based on a control signal, between the voltage of the capacitor and at least one supply voltage including or otherwise developed using the source voltage to provide power to the always-on load during the low power mode. The select control circuit provides the control signal to control power provided to the always-on load during the low power mode.

Recycling capacitance energy from active mode to low power mode

An electronic device includes a digital circuit, a power delivery subsystem configured to provide a supply voltage and a body-biasing voltage to the digital circuit, and a controller a controller coupled to the power delivery subsystem. The controller is configured to determine a process parameter for the electronic device, determine a current temperature parameter for the electronic device, concurrently determine a first coarse-grain level for the supply voltage and a second coarse-grain level for the body-biasing voltage based on the process parameter, the current temperature parameter, and a frequency of a clock signal to be supplied to the digital circuit, and to determine a fine-grain level for the supply voltage.

Closed-loop adaptive voltage, body-biasing and frequency scaling

A technique for managing power in an embedded processing system includes generating a workload model for the embedded processing system in response to a control signal, an event status signal, and a reference clock signal. The control signal is received from a peripheral device. The event status signal is received from an event processor configured to control execution of tasks by a processor core of the embedded processing system. The technique includes providing power configuration information to the processor core. The power configuration information corresponds to an operating point selected based on the control signal, the event status signal, the reference clock signal, the workload model, and a predetermined energy model.

Event-based power manager

A level shifter circuit is described herein for shifting a signal from a first voltage domain to a second voltage domain. The level shifter circuit includes two current paths between a supply terminal of the first voltage domain and a supply terminal of the second voltage domain. The first and second current paths each include a differential transistor that receives a signal from a pulse generator in a first voltage domain. The pulse generator provides pulses to the differential transistors based on an input signal to be translated to the second voltage domain. The level shifter includes a latch circuit in the second voltage domain that includes two inputs where each input is biased at a node of one of the current paths. Each current path includes a bias transistor whose control terminal receives a compensated biasing voltage for biasing the bias transistor. The compensated biasing voltage is compensated to account for drive strength variation of at least one transistor in each current path.

Level shifter circuit with transistor drive strength variation compensation

One example discloses an apparatus for charge recycling between a first power-domain operating at a first voltage and a second power-domain operating at a second voltage, including: a first power-delivery circuit configured to supply the first voltage to the first power-domain; and a second power-delivery circuit coupled to receive power from both the first power-delivery circuit and the first power-domain; wherein the second power-delivery circuit is configured to supply the second voltage to the second power-domain.

Apparatus for charge recycling

An intelligent interrupt distributor balances interrupts (workload) in a highly parallelized system. The intelligent interrupt distributor distributes the interrupts between the processor cores. This allows lowering of voltage and frequency of individual processors and ensures that the overall system power consumption is reduced.

Multiprocessor system with interrupt distributor

An integrated circuit comprising a set of data processing units including a first data processing unit and at least one second data processing unit operable at variable frequencies is disclosed. The integrated circuit further includes an instruction scheduler adapted to evaluate data dependencies between individual instructions in a received plurality of instructions and assign the instructions to the first data processing unit and the at least one second data processing unit for parallel execution in accordance with said data dependencies. The integrated circuit is operable in a first power mode and a second power mode. The second power mode is a reduced power mode compared to the first power mode and is adapted to adjust the operating frequency of the first data processing unit and the at least one second data processing unit in the second power mode as a function of the evaluated data dependencies.

Parallel execution of instructions in processing units and adjusting power mode based on monitored data dependency over a period of time

This invention provides a clock control circuit, which can be added to any pipeline-processor to solve timing problems arising from variations due to process outcome and environmental conditions. Critical instructions are detected (instructions that exercise critical paths) in conjunction with environmental sensing (such as process, temperature and voltage). This information is used to control cycle stealing.

Cycle stealing when a process or environmental variation is detected and a processing instruction is using a critical path

Embodiments of a processing architecture are described. The architecture includes a fetch unit for fetching instructions from a data bus. A scheduler receives data from the fetch unit and creates a schedule allocates the data and schedule to a plurality of computational units. The scheduler also modifies voltage and frequency settings of the processing architecture to optimize power consumption and throughput of the system. The computational units include control units and execute units. The control units receive and decode the instructions and send the decoded instructions to execute units. The execute units then execute the instructions according to relevant software.

Energy efficient microprocessor platform based on instructional level parallelism

The invention provides a clock select circuit and method which uses feedback arrangements between latches in different branches, with each branch for coupling an associated clock signal to the circuit output. An override circuit is provided in one of the feedback arrangements for preventing a latching delay in that feedback arrangement. This enables rapid switching between clocks in both directions.

Growing community of inventors

Eindhoven, Netherlands

Hamed Fatemi