Apple Inc.

Ajay Bhatia

Vivekanandan Venugopal

Michael R Seningen

Michael A Dreesen

Mohamed Hassan Abu-Rahma

Greg M Hess

Bo Tang

Suparn Vats

Yildiz Sinangil

Shahzad Nazar

Amrinder S Barn

Raghava Rao V Denduluri

Suresh Balasubramanian

Qi Ye

Aravind Kandala

Gopinath Venkatesh

Teng Wang

Victor Zyuban

Jaroslav Raszka

Zhao Wang

Hang Huang

Bharan Giridhar

Wenhao Li

Mayur V Joshi

Shingo Suzuki

Daniel C Chow

Steven C Sullivan

Naveen Javarappa

Huaimin Li

Sheela R Shreedharan

Raymond Chang

Sneha Sindhu Matam

Hemangi U Gajjewar

Anshul Y Mehta

Siddhesh Gaiki

A serializer circuit for a computer system is disclosed. The serializer circuit may receive multiple asynchronous signals, and generate a particular serialized signal by blocking transitions of a particular asynchronous signal that corresponds to the particular serialized signal in response to a detection of a transition of at least one of the remaining asynchronous signals that does not include the particular asynchronous signal. The serializer circuit may also generate a different serialized signal by blocking transitions of a different asynchronous signal that corresponds to the different serialized signal in response to a detection of a transition of at least one of the remaining asynchronous signals that does not include the different asynchronous signal. A digital circuit may generate a particular output signal using the particular serialized signal, and generate a different output signal using the different serialized signal.

Clock-free serialization of asynchronous signals

A compute-memory circuit included in a computer system includes multiple data storage cells and multiplier circuits. The data storage cells store weight values associated with a first operand. The multiplier circuits are coupled to a global bit line and receive the weight values via local bit lines coupled to the data storage cells. Using the received weight values and activation signals indicative of a second operand, the multiplier circuits modify a voltage level of the global bit line. The resultant voltage level on the global bit line is indicative of a product of the first and second operands, and can be converted to a digital value using an analog-to-digital converter circuit.

Performing multiple bit computation and convolution in memory

A compute-memory circuit included in a computer system may include multiple compute data storage cells coupled to a compute bit line via respective capacitors. The compute data storage cells may store respective bits of a weight value. During a multiply operation, an operand may be used to generate a voltage level on a compute word line that is used to store respective amounts of charge on the capacitors, which are coupled to the compute bit line. The voltage on the compute bit line may be converted into multiple bits whose value is indicative of a product of the operand and the weight value.

Memory bit cell for in-memory computation

A memory circuit including a storage array circuit and a control circuit is disclosed. The control circuit may select a particular row of storage circuits in the storage array circuit and activate a first-stage circuit in a particular storage circuit in the particular row by coupling the first-stage circuit to a power supply node. At a later point in time, the control circuit may initiate a transfer of write data from the first-stage circuit to a second-stage circuit in the particular storage circuit and de-activate the first-stage circuit by de-coupling the first-stage circuit from the power supply node. Additionally, the control circuit may also de-select the particular row, at a different point in time, and maintain a de-activated state of the first-stage circuit until a subsequent selection of the particular row.

Power loss reduction in data storage arrays

An apparatus includes a control circuit configured to selectively activate, based on an operating mode signal, either a local clock signal or a pulse signal. The apparatus further includes a data storage circuit that is coupled to a data signal, the local clock signal, and the pulse signal. The data storage circuit may be configured to sample the data signal using the local clock signal during a first operating mode, and to sample the data signal using the pulse signal during a second operating mode.

Hybrid pulse/two-stage data latch

A power detection circuit for detecting powering down of a voltage domain in an integrated circuit is disclosed. The power detection circuit is placed in or near the voltage domain in the integrated circuit to provide power detection on the integrated circuit. The power detection circuit detects powering down of the voltage domain to provide an isolation enable signal to another voltage domain that interfaces with the powering down voltage domain. The isolation enable signal may be used by an isolation cell coupled to the non-powering down voltage domain to prevent corrupted logic being received from the powering down voltage domain.

Power down detection for non-destructive isolation signal generation

A flip-flop with glitch protection is disclosed. The flip-flop includes a differential amplifier circuit that generates amplifier output signals based on an input data and clock signals and precharges a true data node when a clock signal is inactive. A latch circuit is coupled to the differential amplifier and includes a latch node. Responsive to a current value of the input data signal having a first logic state, the latch node is set at a logic value equivalent to the precharged value during an active phase of the clock signal. Responsive to the current value of the input data signal having a second logic state complementary to the first, during the active phase of the clock signal, the latch circuit causes the latch node to be set to a logic value complementary to the precharged value, using the clock signal and the current value of the input data signal.

Flip-flop circuit with glitch protection

Systems, apparatuses, and methods for implementing low voltage clock swing sequential circuits are described. An input signal is coupled to the gates of a first P-type transistor and a first N-type transistor of a first transistor stack. A low voltage swing clock signal is coupled to the gate of a second N-type transistor of the first transistor stack. An inverse of the input signal is coupled to the gates of a second P-type transistor and a third N-type transistor of a second transistor stack. The low-swing clock is coupled to the gate of a fourth N-type transistor of the second transistor stack. A first end of one or more enabling P-Type transistors with gates coupled to the low-swing clock is coupled to the first P-type transistor's drain, and a second end of the one or more enabling P-Type transistors is coupled to the second P-type transistor's drain.

Low voltage clock swing tolerant sequential circuits for dynamic power savings

A system and method for efficiently storing and driving data between pipeline stages. In various embodiments, a flip-flop circuit includes a bypass circuit, which is a tri-state inverter, and the bypass circuit receives a clock signal and a version of a data signal. When the clock signal received by the flip-flop circuit is asserted, the output of the bypass circuit is sent as the output of the flip-flop circuit. In one example, the version of the data signal received by the bypass circuit is the data signal. In another example, the version of the data signal received by the bypass circuit is the output of a master latch. Although the output of the master latch is pre-charged, when the clock is asserted, each of a late arriving rising and falling data transition are included in the critical path of the flip-flop circuit.

Semi dynamic flop and single stage pulse flop with shadow latch and transparency on both input data edges

Techniques are disclosed relating to level-shifting circuitry and time borrowing across voltage domains. In some embodiments, sense amplifier circuitry generates, based on an input signal at a first voltage level, an output signal at a second, different voltage level. Pulse circuitry may generate a pulse signal in response to an active clock edge of a clock signal that is input to the sense amplifier circuitry. Initial resolution circuitry may drive the output signal of the sense amplifier circuitry to match the value of the input signal during the pulse signal. Secondary resolution circuitry may maintain a current value of the output signal after expiration of the pulse signal. This may allow the input signal to change during the pulse, e.g., to enable time borrowing by upstream circuitry.

Level-shifting transparent window sense amplifier

Systems, apparatuses, and methods for implementing low-power flip-flops with balanced clock-to-Q delay are described. A flip-flop includes a primary latch, an upper secondary latch, and a lower secondary latch. The primary latch transmits a data value from an input port to a first node when transparent. The upper secondary latch pulls up a second node when transparent and when the first node is equal to a first value. The second node is a prebuffered data output of the flip-flop. The lower secondary latch pulls down the second node when transparent and when the first node is equal to a second value different from the first value. To ensure the flip-flop has a balanced clock-to-Q delay, a first set of clock signals coupled to transistor gates of the primary latch are delayed with respect to a second set of clock signals coupled to transistor gates of the upper secondary latch.

Low power flip-flop with balanced clock-to-Q delay

Growing community of inventors

Saratoga, CA, United States of America

Ajay Bhatia