Apparatus having dice to perform refresh operations

This application is a divisional of U.S. application Ser. No. 15/431,383, filed Feb. 13, 2017, which is a continuation of U.S. application Ser. No. 14/714,962, filed May 18, 2015, now issued as U.S. Pat. No. 9,570,142, all of which are incorporated herein by reference in their entireties.

Some semiconductor devices, such as a DRAM (Dynamic Random Access Memory), perform a refresh operation to restore charge in memory cells to maintain the stored state of logical data. In general, the refresh operation is performed on a periodic basis or on a command basis with regard to a single die of DRAM. When a plurality of dice are involved, operations are more complex, such that current consumption and stability become more difficult to control.

According to a first aspect of the invention, a apparatus includes an interface chip including an oscillator producing an original clock signal, a first memory chip including a first terminal configured to receive a first clock signal in response to the original clock signal, a second terminal supplied with a first control signal, and first memory cells that are subject to a data refresh operation in response to the first clock signal when the first control single is active, and a second memory chip including a third terminal configured to receive a second clock signal in response to the original clock signal, a fourth terminal supplied with a second control signal, and second memory cells that are subject to a data refresh operation in response to the second clock signal when the second control signal is active.

According to a second aspect of the invention, a apparatus includes a chip stack structure including first and second memory chips stacked over each other, and an interface chip coupled to the chip stack structure. The first memory chip includes first and second terminals and a plurality of first memory cell. The second memory chip includes third, fourth and fifth terminals. The first and second memory chips are stacked over each other such that the second terminal of the first memory chip is electrically connected to the fifth terminal of the second memory chip. The interface chip includes sixth, seventh, and eighth terminals, and is coupled to the chip stack structure such that the sixth, seventh and eighth terminals of the interface chip are electrically connected to the third, fourth and fifth terminals of the second memory chips, the second terminal is coupled to a first command decoder circuit of the first memory chip, and the fourth terminal is coupled to a second command decoder circuit of the second memory chip. The first and second command decoder circuits work independently from each other.

According to a third aspect of the invention, an apparatus includes an interface chip that includes an oscillator circuit to provide an oscillator signal, a first memory chip on which a first refresh operation is performed in response to a first refresh enable signal supplied at a first node and a first oscillator enable signal supplied at a second node, a second memory chip on which a second refresh operation is performed in response to a second refresh enable signal supplied at a third node and a second oscillator enable signal supplied at a fourth node. The first and third nodes are configured to receive from the interface chip the first and second refresh enable signals independently of each other. The second and fourth nodes are configured to receive, based on the oscillator signal, the first and second oscillator enable signals in different timings from each other so that the first and second refresh operations are performed in different timings from each other even when the first and third nodes receive the first and second refresh enable signals simultaneously.

FIG. 1 illustrates an example of a high bandwidth memory according to a first embodiment.

FIG. 2 illustrates an example structure having a high bandwidth memory and a graphics processing unit according to the first embodiment.

FIG. 3 illustrates an example of the circuitry of the interface die according to the first embodiment.

FIG. 4 illustrates an example of a high bandwidth memory according to a second embodiment.

FIG. 5 illustrates an example of self-refresh timing according to the second embodiment.

FIG. 6 illustrates another example of self-refresh timing according to the second embodiment.

FIG. 7 illustrates an example of a high bandwidth memory according to the first embodiment.

FIG. 8 illustrates a third example of a high bandwidth memory according to a third embodiment.

FIG. 9 illustrates an example of a self-refresh timing according to the first embodiment.

FIG. 10 illustrates an example of a core die according to the first embodiment.

FIG. 11 illustrates an example of self-refresh waveforms in a core die according to the first embodiment.

FIG. 12 illustrates a fourth example of a high bandwidth memory according to a fourth embodiment.

FIG. 13 illustrates a fifth example of a high bandwidth memory according to a fifth embodiment.

The description that follows includes illustrative an apparatus including semiconductor devices (circuits, systems, and the like) and processes (e.g., timing, waveforms, and the like) that embody the disclosed subject matter. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide an understanding of various embodiments of the subject matter.

FIG. 1 illustrates an example of a high bandwidth memory (HBM) according to a first embodiment. The HBM 100 may include an interface die 101 , a stack of core dice 102 over the interface die 101 , through-silicon-vias (TSVs) 103 to connect the core dice 102 with each other, and bump pads 104 , and the HBM 100 may include a high band width memory explained below for example by referring to FIGS. 7, 8, 12, and 13 .

In some embodiments, the bump pads 104 may include micro bumps 104 , and may receive signals from a host controller (not shown). The interface die 101 may buffer signals from the bumps 104 to the TSVs 103 . In some embodiments, the core dice 102 may include memory cells, while the interface die 101 may not include any memory cells. In some embodiments, the core dice 102 may include one or more command decoders (such as a command decoder 116 as shown in FIG. 7 ).

In some embodiments, the interface die 101 may be divided into some independent channels. These channels may be independent of one another. The HBM 100 may have a wide data interface that may perform an operation with a high-speed and low-power-consumption. A self-refresh operation may be performed in the HBM 100 .

FIG. 2 illustrates an example structure 200 including a HBM 100 and a Graphics Processing Unit (GPU) 105 according to the first embodiment. In some embodiments, a re-driver 106 may be disposed between the HBM 100 and the GPU 105 to drive a signal. The structure 200 may be called as a semiconductor device, an apparatus, or a system, that is accessed via a connection portion by a controller (not shown) or a processor (not shown). The connecting portion may be a bump, a ball, or a solder ball. FIG. 3 illustrates an example of the circuitry of the interface die 101 according to the first embodiment. In some embodiments, the circuitry of the interface die 101 may include a memory Built-In-Self-Test (BIST) circuit 107 to test the core die 102 , internal circuits 108 (as shown in FIG. 4 for example), and an input buffer 109 to receive signals from a corresponding micro bumps 104 and to convey the signals to corresponding TSVs 103 .

FIG. 4 illustrates a first example of a HBM 400 according to a second embodiment. As shown in FIG. 4 , the HBM 400 may include an interface die 401 , and a stack of core dice 402 (such as core die 402 A and core die 402 B), which are connected to each other by TSVs 403 .

In some embodiments, the interface die 401 may include bumps 404 (such as bump_A and bump_B), and internal circuits 408 (which may have receivers (Rx) 410 , transmitters (Tx) 411 , transceivers (Tx Rx) 412 , logic 413 , and buffers 414 ).

In some embodiments, the core die 402 A may include memory cells located in a DRAM memory array 415 , a command decoder 416 , a self refresh oscillator 417 , and a data controller 418 . Similarly, the core die 402 B may include memory cells located in a DRAM memory array 415 , a command decoder 416 , a self refresh oscillator 417 , and a data controller 418 . The core die 402 A and the core die 402 B may operate on commands respectively, and may thus work independently from each other.

FIG. 5 illustrates an example of self-refresh timing when both the core A and the core B as shown in FIG. 4 are in a self-refresh state according to the second embodiment. FIG. 6 illustrate another example of self-refresh timing when the core A as shown in FIG. 4 is in a self-refresh state and the core B as shown in FIG. 4 is not in a self-refresh state according to the second embodiment. In FIG. 5 and FIG. 6 , self_osc represents a Self Refresh Oscillator signal, ACT represents an Active Command (which can be provided by a GPU), WR represents a WRITE signal, and RD represents a READ signal.

A DRAM die 102 A has a self refresh function and a DRAM die 102 B also has a refresh function. The refresh function may be known as a self refresh mode of a DRAM. In self-refresh mode, the DRAM may automatically refresh the memory cell data, and all banks in the DRAM may be activated to refresh the memory cell data of the banks.

In the second embodiment, refresh operations of the two dice 102 A and 102 B are performed independently of each other, and thus those two refresh operation may sometimes occur simultaneously, which may be indicated as same timing in FIG. 5 . Therefore, consumption current in the same timing of refresh operations of two dice 102 A and 102 B may become higher than consumption current in different timing of refresh operations. So to speak, peak current in the second embodiment HBM may sometimes become high in the same timing refresh operations. The DRAM dice of HMB may consume a large amount of power at the same time, and thus may induce power noise, possibly affecting the integrity of the self refresh function.

FIG. 7 illustrates an example of a HBM 700 according to the first embodiment. The explanations above may be applicable and/or identical at least in part to the first embodiment, and thus for simplicity, some common explanations are omitted. For example, the configurations shown above may be referred here according to the first embodiment.

As shown in FIG. 7 , the HBM 700 may include an interface die 701 , and a stack of core dice 702 (includes at least two dice such as a core die 702 A and a core die 702 B), which are connected to each other by TSVs 703 .

In some embodiments, the interface die 701 may include bumps 704 (such as a bump_A and a bump_B), and internal circuits 708 . The internal circuits 708 of the interface die 701 may include a self refresh oscillator 717 , receivers 710 , transmitters 711 , transceivers 712 , logic 713 , and buffers 714 .

Here, in some embodiments, the self refresh oscillator 717 may be located in the interface die 701 , and may work in common with both the core die 702 A and the core die 702 B. In contrast, a self refresh oscillator 427 is not located in the interface die 401 and is located in each of dice 401 A and 402 B in FIG. 4 Thus, in the second embodiment, refresh operation may work independently of each other. According to the first embodiment, the self refresh oscillator 717 is configured common to stack of die including dice 702 A and 702 B.

In some embodiments, the self-refresh oscillator 717 may be activated according to a reset signal when the HBM 700 is not in a reset state for example.

In some embodiments, the core die 702 A may include memory cells located in a DRAM memory array 715 , a command decoder 716 , a data controller 718 , and a delay adjust circuit 719 A. Similarly, the core die 702 B may include memory cells located in a DRAM memory array 715 , a command decoder 716 , a data controller 718 , and a delay adjust circuit 719 B. The core die 702 A and the core die 702 B may operate on commands respectively, and thus may work independently of each other.

Here, according to the first embodiment, the delay adjust circuit 719 A of the core die 702 A and the delay adjust circuit 719 B of the core die 702 B are coupled in series to each other. In such an arrangement, a delay amount of the delay adjust circuit 719 A may be configured to be the same as a delay amount of the delay adjust circuit 719 B, and the delay adjust circuits 719 A and 719 B may be configured to be identical to each other.

According to the first embodiment, the refresh timing of the channels may be different due to the delay adjust circuits 719 (such as the delay adjust circuit 719 A and the delay adjust circuit 719 B), and thus a peak current consumption of the associated semiconductor device may be reduced.

FIG. 8 illustrates a third example of a HBM 800 according to a third embodiment. As shown in FIG. 8 , the HBM 800 may include an interface die 801 , and a stack of core dice 802 (includes at least two dice such as a core die 802 A and a core die 802 B), which are connected to each other by TSVs 803 . The explanations above may be applicable and/or identical at least in part to the third embodiment, and thus for simplicity, some common explanations are omitted. For example, the configurations shown above may be referred here according to the third embodiment.

In some embodiments, the interface die 801 may include bumps (such as a bump_A and a bump_B), and internal circuits 808 . The internal circuits of the interface die 801 may include a self refresh oscillator 817 , receivers 810 , transmitters 811 , transceivers 812 , logic 813 , and buffers 814 .

In some embodiments, the core die 802 A may include memory cells located in a DRAM memory array 815 , a command decoder 816 , a data controller 818 , and a delay adjust circuit 819 A. Similarly, the core die 802 B may include memory cells located in a DRAM memory array 815 , a command decoder 816 , and a data controller 818 , and a delay adjust circuit 819 B. The core die 802 A and the core die 802 B may operate on commands respectively, and thus may work independently of each other.

According to the second embodiment, the delay adjust circuit 819 A of the core die 802 A and the delay adjust circuit 819 B of the core die 802 B are coupled in parallel to each other. In such an arrangement, a delay amount of the delay adjust circuit 819 A may be different from a delay amount of the delay adjust circuit 819 B, and the delay adjust circuits 819 A and 819 B may be configured to be different from each other.

According to the second embodiment, the refresh timing of the channels may be different due to the delay adjust circuits 819 (such as the delay adjust circuit 819 A and the delay adjust circuit 819 B), and thus a peak current consumption of the associated semiconductor device may be reduced.

FIG. 9 illustrates an example of a-self refresh timing according to the first embodiment. The timing in FIG. 9 may be also referred according to other embodiments.

As shown in FIG. 9 , two self refresh operations to the core die 102 A and the core die 102 B can be performed at different timing values due to the delay adjust circuits 119 A and 119 B for example.

According to the embodiments, two self oscillator singles SELF_OSC_A and SELF_OSC_B are commonly generated from one self oscillator signal SELF_OSC_IF and those two self oscillator singles SELF_OSC_A and SELF_OSC_B are triggered and activated to be high in different timing. Thus, simultaneous self-refresh operations of dice in stack can be prevented, and peak current in a self-refresh operation of HMB can be mitigated and become lower than the second embodiment.

FIG. 10 illustrates an example of a core die 102 according to the first embodiment. The configuration shown in FIG. 10 may be also referred according to other embodiments.

As shown in FIG. 10 , the core die 102 (e.g., the core die 102 A as shown in FIG. 8 ) may include a command decoder 116 , and a delay adjust circuit 119 . When any channel is in a self-refresh mode, a self oscillator signal may be driven to the core die 102 . The delay adjust circuit 119 in the core die 102 may change a self refresh timing in each channel of the core die 102 .

FIG. 11 illustrates an example of refresh waveforms in a core die 102 according to the first embodiment. The waveform in FIG. 11 may be also referred according to other embodiments.

In FIG. 11 , “self_en” represents a Self Refresh Entry signal, which may be provided by a GPU. “self_exit” represents a Self Refresh Exit signal, which may be provided by a GPU. “self_st” represents a Self Refresh State signal, which may go up from the “self_en” and go down from the “self_en”. “self_osc” represents a Self Refresh Oscillator signal. “ref_go” represents a Refresh Start signal. “soak” represents a signal that is generated from a rising edge of the “ref_go”. The “soak” signal may come from a memory array region, and may indicate that it is ok to finish the refresh operation. “ref_state” represents a Refresh Command (State) signal, which may go to a memory array region. The refresh operation may be started from a rising edge of the “ref_state”, and may be finished at a falling edge of the “ref_state”.

FIG. 12 illustrates an example of a HBM 1200 according to a fourth embodiment. The explanations above may be applicable and/or identical at least in part to the fourth embodiment, and thus for simplicity, some common explanations are omitted. For example, the configurations shown above may be referred here according to the fourth embodiment.

As shown in FIG. 12 , the HBM 1200 may include an interface die 1201 , and a stack of core dice 1202 (such as a core die 1202 A and a core die 1202 B), which are connected to each other by TSVs 1203 .

In some embodiments, the interface die 1201 may include bumps (such as a bump_A and a bump_B), and internal circuits 1208 . The internal circuits 1208 of the interface die 1201 may include a self refresh oscillator 1217 , a delay adjust circuit 1219 A, a delay adjust circuit 1219 B, receivers 1210 , transmitters 1211 , transceivers 1212 , logic 1213 , and buffers 1214 . The delay adjust circuit 1219 A may receive a “self_osc” signal from the self-refresh oscillator 1217 for example.

Here, according to the fourth embodiment, a delay adjust circuit 1219 A and a delay adjust circuit 1219 B are located in the interface die 1201 . In contrast, a delay adjust circuit 1219 A and a delay adjust circuit 1219 B are not located in the interface die 1201 , and are located in core dice, respectively.

According to the fourth embodiment, the delay adjust circuit 1219 A may be coupled to the core die 1202 A. The delay adjust circuit 1219 B may be coupled between the delay adjust circuit 1219 A and the core die 1202 B, and thus the delay adjust circuit 1219 B may be coupled in series with the delay adjust circuit 1219 A. In such an arrangement, the circuit layout area in each core die (such as the core die 1202 A and the core die 1202 B) may be reduced.

In some embodiments, the core die 1202 A may include memory cells located in a DRAM memory array 1215 , a command decoder 1216 , and a data controller 1218 . Similarly, the core die 1202 B may include memory cells located in a DRAM memory array 1215 , a command decoder 1216 , and a data controller 1218 . The core die 1202 A and the core die 1202 B may operate on command respectively, and thus may work independently of each other.

FIG. 13 illustrates a fifth example of a HBM 1300 according to a fifth embodiment. The explanations above may be applicable and/or identical at least in part to the fifth embodiment, and thus for simplicity, some common explanations are omitted. For example, the configurations shown above may be referred here according to the fifth embodiment.

As shown in FIG. 13 , the HBM 1300 may include an interface die 1301 , and a stack of core dice 1302 (such as a core die 1302 A and a core die 1302 B), which are connected to each other by TSVs 1303 .

In some embodiments, the interface die 1301 may include bumps (such as a bump_A and a bump_B), and internal circuits 1308 . The internal circuits 1308 of the interface die 1301 may include a self refresh oscillator 1317 , a delay adjust circuit 1319 A, a delay adjust circuit 1319 B, receivers 1310 , transmitters 1311 , transceivers 1312 , logic 1313 , and buffers 1314 .

Here, according to the fourth embodiment, a delay adjust circuit 1319 A and a delay adjust circuit 1319 B are located in the interface die 1301 . In contrast, a delay adjust circuit 1319 A and a delay adjust circuit 1319 B are not located in the interface die 1301 , and are located in core dice, respectively.

According to the fourth embodiment, the delay adjust circuit 1319 A and the delay adjust circuit 1319 B are coupled in parallel to each other, and may receive a “self_osc” signal from the self-refresh oscillator 1317 for example. The delay adjust circuit 1319 A is coupled between the self refresh oscillator 1317 and the core die 1302 A. The delay adjust circuit 1319 B is coupled between the self-refresh oscillator 1317 and the core die 1302 B.

In some embodiments, the core die 1302 A may include memory cells located in a DRAM memory array 1315 , a command decoder 1316 , and a data controller 1318 . Similarly, the core die 1302 B may include memory cells located in a DRAM memory array 1315 , a command decoder 1316 , and a data controller 1318 . The core dice 1302 A and 1302 B may operate on command respectively, and thus may work independently of each other.

In some embodiments, the delay adjust circuit 1319 A of the interface die 1301 may be coupled to the DRAM memory array 1315 of the core die 1302 A via elements (such as a transmitter 111 of the interface die 1301 , a TSV 1303 , a receiver 1310 of the core die 1302 A, and logic 1313 of the core die 1302 A in series). Similarly, the delay adjust circuit 1319 B of the interface die 1301 may be coupled to the DRAM memory array 1315 of the core die 1302 B via elements (such as a transmitter 1311 of the interface die 1301 , a TSV 1303 , a receiver 1310 of the core die 1302 A, and logic 1313 of the core die 1302 A in series).

In such an arrangement, an apparatus may only have one or more self refresh oscillators in an interface die, and the power consumption of the apparatus can be reduced.

The Abstract of the Disclosure is provided to comply with 37 C.F.R. § 1.72(b), requiring an abstract that will allow the reader to quickly ascertain the nature of the technical disclosure. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims. In addition, in the foregoing Detailed Description, it can be seen that various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. Thus the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separate embodiment.