EXPLOITING PCM WRITE ASYMMETRIES TO ACCELERATE WRITE
First Claim
1. A phase-change memory (PCM) chip comprising a circuit, wherein the circuit comprises:
- a buffer, the buffer configured to store one or more bits and to output a buffer signal;
a finite states machine (FSM), the FSM configured to provide an output control, a first write-region control, and a second write-region control, wherein the first and second write-region controls specify which write-region is scheduled;
a first AND gate, the first AND gate configured to receive the output control, the first write-region control, and an inverted buffer signal, wherein the inverted buffer signal is an inversion of the buffer signal output by the buffer;
a second AND gate, wherein the second AND gate is configured to receive the buffer signal, the second write-region control, and an inversion of the output control;
an OR gate, wherein the OR gate is configured to receive an output signal of the first AND gate and an output signal of the second AND gate; and
a switch, the switch configured to provide the buffer signal to a write driver according to an output signal of the OR gate, wherein the write driver writes one or more bits stored in the buffer in the target cache line;
wherein the circuit is configured to schedule writing requests in at least two stages, wherein a first stage comprises writing all bits of zero in a target cache line and a second stage comprises writing all bits of one in the target cache line.
1 Assignment
0 Petitions
Accused Products
Abstract
To improve the write performance of PCM, the disclosed technology, in certain embodiments, provides a new write scheme, referred to herein as two-stage-write, which leverages the speed and power asymmetries of writing a zero bit and a one bit. Writing a data block to PCM is divided into two separated stages, i.e., write-0 stage and write-1 stage. Without violating power constraints, during the write-0 stage, all zero bits in this data block are written to PCM at an accelerated speed; during the write-1 stage, all one bits are written to PCM, with more bits being written concurrently. In certain embodiments, the disclosed technology provides a new coding scheme to improve the speed of the write-1 stage by further increasing the number of bits that can be written to PCM in parallel.
4 Citations
14 Claims
-
1. A phase-change memory (PCM) chip comprising a circuit, wherein the circuit comprises:
-
a buffer, the buffer configured to store one or more bits and to output a buffer signal; a finite states machine (FSM), the FSM configured to provide an output control, a first write-region control, and a second write-region control, wherein the first and second write-region controls specify which write-region is scheduled; a first AND gate, the first AND gate configured to receive the output control, the first write-region control, and an inverted buffer signal, wherein the inverted buffer signal is an inversion of the buffer signal output by the buffer; a second AND gate, wherein the second AND gate is configured to receive the buffer signal, the second write-region control, and an inversion of the output control; an OR gate, wherein the OR gate is configured to receive an output signal of the first AND gate and an output signal of the second AND gate; and a switch, the switch configured to provide the buffer signal to a write driver according to an output signal of the OR gate, wherein the write driver writes one or more bits stored in the buffer in the target cache line; wherein the circuit is configured to schedule writing requests in at least two stages, wherein a first stage comprises writing all bits of zero in a target cache line and a second stage comprises writing all bits of one in the target cache line. - View Dependent Claims (2, 3, 4, 5, 6, 7)
-
- 8. A phase-change memory (PCM) chip comprising a circuit, the circuit configured to schedule writing requests in at least two stages, wherein a first stage of the at least two stages comprises writing all bits of zero in a target cache line and a second stage of the at least two stages comprises writing all bits of one in the target cache line.
Specification