CN100485927C - 3D storage - Google Patents

Abstract

The invention improves three-dimensional memory (3D-4) further and uses the integration capability to increase speed of 3D-M, rate of finished products and programmable capability. 3D-M uses the sense amplifier (S/A), full reading mode and self-timing to raise the reading speed. 3D-M capacity in unit array can be greatly increased by using S/A and full reading mode so as to enhance the integration capability between 3D-M and the circuit on the substrate. RAM embedded in the substrate can be as the catch of 3D-M data or ROM embedded in the substrate can be as error correction data and/or upgrading code of 3D-M data.

Description

Three-dimensional storage

Technical field

The present invention relates to integrated circuit fields, or rather, relate to the design of three-dimensional storage.

Background technology

Three dimensional integrated circuits (abbreviating 3D-IC as) is stacked in one or more three dimensional integrated circuits layers (abbreviating the 3D-IC layer as) on the substrate on the direction perpendicular to substrate mutually.3D-IC can have multiple function, as analog functuion, digital function, memory function etc.Because memory has error correcting capability, it can tolerate bigger defect concentration; And it is low in energy consumption, does not have heat dissipation problem, so that memory is particularly suitable for is three-dimensional integrated.

Three-dimensional storage (3-dimensional memory abbreviates 3D-M as) is stacked in one or more accumulation layers on the substrate circuitry on the direction perpendicular to substrate mutually.Shown in Figure 1A, 3D-M contains at least one and is stacked and placed on three-dimensional accumulation layer 100 on the Semiconductor substrate 0s, many address selection lines (comprising word line 20a and bit line 30a) and a plurality of three-dimensional storage element, i.e. 3D-M unit (1aa...) are arranged on each three-dimensional accumulation layer (as 100).A plurality of transistors are arranged on the substrate 0s.Contact through hole (20av, 30av...) provide electrical connection for address selection line (20a, 30a...) and substrate circuitry.3D-M can be divided into three-dimensional random access memory (3D-RAM) and 3 D ROM (3D-ROM).The circuit of 3D-RAM unit and conventional RAM unit are similar, and just it generally constitutes (Figure 1B) by thin-film transistor 1t.3D-ROM can be masking film program (3D-MPROM), one-time programming (3D-EPROM) or repeatedly programming (comprising 3D-flash, 3D-MRAM, 3D-FRAM, 3D-OUM etc.) at least.Visible United States Patent (USP) 5,835,396 open files such as grade of its basic structure.It can use as the active element of thin-film transistor (TFT) 1t (Fig. 1 CA, Fig. 1 CB) and/or as the passive component (Fig. 1 DA-Fig. 1 E) of diode 1d.For the 3D-ROM unit that uses TFT, they can contain suspension grid 30fg (Fig. 1 CA) or have vertical-channel 25c (Fig. 1 CB).For the 3D-MPROM unit that uses diode, it contains the 3D-ROM film 22 (comprising accurate conductive membrane) with nonlinear resistance property, and come presentation logic " 1 " with exist (not existing of medium 26 promptly is set) of information opening 24 (being access opening), (Fig. 1 DA), do not exist (existence of medium 26 promptly is set) of information opening 24 comes presentation logic " 0 " (Fig. 1 DB).Here, medium 26 is set is meant medium between address selection line 20a, 30a, whether it exists the value of setting that determines this 3D-ROM unit for the 3D-EPROM that uses diode, can come presentation logic information (Fig. 1 E) by the integrality of anti-fuse 22af.

3D-M has advantages such as low cost, high density, but since its storage element generally constitute by non-single crystalline semiconductor material, so its performance still be difficult to the routine, the semi-conductive solid-state memory of based single crystal is compared.This need be further improved the peripheral circuit of 3D-M, but and make full use of the integration of itself and substrate circuitry, improve 3D-M speed, rate of finished products and programmability.The present invention has done further perfect in these areas to 3D-M.

Goal of the invention

Main purpose of the present invention is further to improve the performance of three-dimensional storage (3D-M).

Another object of the present invention is to improve the speed of 3D-M.

But another object of the present invention is to improve the integration of 3D-M.

Another object of the present invention is to improve the rate of finished products of 3D-M.

Another object of the present invention is to improve the programmability of 3D-M and the upgradability of contained information.

According to these and other purpose, the invention provides multiple improved three-dimensional storage (3D-M).

Summary of the invention

With routine, the semi-conductive solid-state memory of based single crystal compares, the read or write speed of 3D-M unit is slower, this can solve from the angle of circuit design and system design.From the angle of circuit design, can utilize sense amplifier (S/A), full-time course pattern and self-timing to improve its reading rate.Owing to use S/A, produce the required bit-line voltage amplitude of oscillation of logic output very little (～0.1V), so the pairs of bit line charging only needs the short period, this can greatly shorten the time of visiting first; The full-time course pattern is read the data in all storage elements on the word line when a read operation simultaneously, and this can improve bandwidth, and can improve the capacity of 3D-M unit's array; Self-timing can improve the confidence level of reading and cut down the consumption of energy.Electricity programming 3D-M can adopt parallel programmed to improve writing rate.

Angle from system design, can use three-D integrated memory (3-dimensional integrated memory, abbreviating 3DiM as, is 02133943.0 patent application " three-D integrated memory " referring to, application number that submit on September 30th, 2002 by same inventor) time of visiting first of hiding 3D-M.Embedded RAM among the 3DiM (embedded RAM abbreviates eRAM as) can be used as the read-write buffer (cache) of 3D-M.After read operation, the 3D-M data that are latched on the S/A are sent to eRAM by segmentation.Correspondingly, kept a backup of 3D-M data among the eRAM.When data were sought by system from 3DiM, it was sought from eRAM earlier, as " hitting ", then directly reads from eRAM; As " miss ", then from 3D-M, read again.Notice that though the performance of single 3D-M unit still is difficult to compare with conventional storage element, by the system integration, its integrated performance can be compared with conventional memory, in addition better.

But big capacity 3D-M unit matrix is shown the integration that helps improve 3D-M.Can improve the capacity of 3D-M unit's array from several aspects.At first, under the full-time course pattern, because the number of bit of unit array is without any restriction, so the 3D-M array can be designed to a rectangle, its number of bit is greater than the word line number.Secondly, because the positive and negative current ratio of the word line number of unit array 3D-ROM unit when being subject to read operation, so can recently improve the word line number by improving positive and negative electric current.The method of a positive and negative current ratio of raising is to use reads voltage V greatly _RBecause the present invention has used designs such as S/A and full-time course pattern, the reverse biased in the positive and negative current ratio is separated with forward bias: maximum reverse is biased in the threshold voltage V of S/A _TNear the (～0.1V); Forward bias is by V _RDecision.In general, forward bias (as～3V) more than reverse biased (as～0.2V) big.By improving V _RCan greatly improve positive and negative current ratio.Another method that improves positive and negative current ratio is to use two polarization 3D-ROM units: second film of going up of two polarization 3D-ROM units contains different sills, or it has different interfaces with upper/lower electrode.

In order to improve the rate of finished products of 3D-M, can be by the first defective number that directly reduces in the 3D-ROM array of a seamless 3D-ROM.Defective sensitive membrane (comprising 3D-ROM film and adjacent with it hearth electrode and top electrode) in the seamless 3D-ROM unit forms with a kind of " seamless " form, i.e. no figure switch process in the forming process of these films.The another kind of method that improves rate of finished products is used the error correction scheme as error correcting code (ECC) and/or redundant circuit etc.They can correct in the 3D-M array mistake that defectiveness causes.Use the 3D-M array of ECC scheme can use row redundant code (as Hamming code).In redundant circuit, the eROM among the 3DiM can be used for the address and the corresponding error correction data of storage defect position.Redundant circuit can carry out error correction to individual bits mistake, bit line mistake, word line failure.This error correction procedure can also can be carried out (promptly " reading the back " repairs) finishing behind the column decoding and carry out (promptly " when reading " repairs) before data being delivered to eRAM in the eRAM that keeps the 3D-M data backup.Redundant circuit is to utilize 3DiM to improve an example of 3D-M rate of finished products in fact.

3DiM can also provide upgrading ability to the contained software code of 3D-M except being used for improving the rate of finished products of 3D-M, comes storing software upgrading sign indicating number as using above-mentioned word line redundant circuit.Software upgrading can also be used the address transition method.In the address transition method, 3D-M and integrated with it Embedded ROM (embedded ROM abbreviates eROM as) form an independent memory space: contained among the 3D-M is source code, and contained among the eROM is upgrading yard.Simultaneously, the substrate integrated circuit also contains an address conversion block, and it is considered as virtual address with Input Address, and converts thereof into the physical address of above-mentioned independent memory space.If actuating code is used source code, so, this physical address points to 3D-M; If actuating code is used the upgrading sign indicating number, so, this physical address points to eROM.

Description of drawings

Figure 1A is the perspective view of a kind of 3D-M; Figure 1B-Fig. 1 CB represents the transistorized 3D-M of multiple based thin film unit; Fig. 1 DA, Fig. 1 DB represent a logical one and " 0 " 3D-MPROM unit respectively; Fig. 1 E represents a kind of 3D-EPROM unit.

Fig. 2 A-Fig. 2 C represents a kind of circuit symbol, fundamental block diagram and detailed diagram of 3D-M nuclear.

Fig. 3 A-Fig. 3 G describes the circuit block that multiple 3D-M nuclear uses.

Fig. 4 AA-Fig. 4 AD explains the source of the time of visiting first; Fig. 4 BA-Fig. 4 CC provides the design of multiple reference bit lines; Fig. 4 D is a kind of implementation method of data bit line in the 3D-ROM array, umbral position line and timing bit line.

Fig. 5 represents the sequential chart of various signals in a kind of 3D-ROM nuclear,

Fig. 6 A-Fig. 6 G represents multiple 3DcM (cached 3D-M) and reads flow process.

Fig. 7 A-Fig. 7 B represents a kind of 3D-EPROM that adopts parallel programmed; Fig. 7 C represents a kind of 3D-EPROM with external programming power supply.

Fig. 8 AA-Fig. 8 G describes the method for multiple raising unit array capacity.

Fig. 9 AA-Fig. 9 CB describes multiple 3D-M defective.

Figure 10 A-Figure 10 B represents two kinds of seamless 3D-ROM units.

The technological process of the multiple seamless 3D-ROM of Figure 11 AA-Figure 11 E ' expression unit.

Figure 12 A-Figure 12 B is the two kinds of seamless 3D-ROM of standard units.

Figure 13 represents a kind of 3D-M error correction (ECC) circuit.

Figure 14 A-Figure 14 DC represents multiple 3D-M redundant circuit.

Figure 15 A-Figure 15 C represents multiple 3D-M with software upgrading function.

Be easy meter, in this manual,, represent that then on behalf of all, it have the figure of this suffix if figure number lacks due suffix.Refer to Fig. 9 AA-Fig. 9 CB as Fig. 9; Fig. 9 C refers to Fig. 9 CA-Fig. 9 CB.

Embodiment

1. read or write speed

This section is example with 3D-ROM, is purpose to improve read or write speed, to the circuit design of 3D-M transistor level, especially to the design of 3D-M nuclear, 3DcM (cached3D-M) and programmed circuit, has done further perfect.Here, 3D-M nuclear is meant 3D-M array and the most basic peripheral circuit that the 3D-M data can be read.In order to improve reading rate,, preferably can use sense amplifier and full-time course pattern, and take self-timing from the angle of circuit; From the angle of system, the time of visiting first that preferably can utilize the eRAM among the 3DcM to hide 3D-M.Correspondingly, though the performance of single 3D-M unit still be difficult to compare with conventional storage element, yet by the system integration, its collective's performance can be compared with conventional memory, even better.In order to improve writing rate, preferably use parallel programmed.

A.3D-M examine

What Fig. 2 A represented 3D-M nuclear 0 reads the IO port.3D-M nuclear comprises 3D-M array and the most basic peripheral circuit thereof.Its input signal comprises row address AS2 and reads enabling signal RD4 that output signal comprises dateout DO8 and ready for data signal RY6.Here do not draw and write the I/O port.

Fig. 2 B is a kind of fundamental block diagram of 3D-ROM nuclear 0.It contains a 3D-ROM array 0A, sense amplifier (S/A) piece 18, turnover voltage (V _M) produce circuit block 14, row decoder 12, bit line and make idle circuit piece 18 ', bias circuit block 16 and address register 121.Wherein, 3D-ROM unit's array 0A contains N _WLBar word line (20c...) and N _BLBit lines (30c...).At word line and bit line infall then presentation logic " 1 " of diode is arranged, diode-less is presentation logic " 0 " then.In order with later the reference bit lines of introducing (comprising regularly bit line and umbral position line etc.) to be distinguished, the bit line 30a-30d that will deposit valid data here is called data bit line.S/A piece 18 zooms into a logical signal 8 with the little analog signal on the bit line, and it is also only just worked when SE5 is high by S/A enable signal SE5 control.V _MProduce circuit block 14 and produce a turnover voltage V _MWhen the bias voltage of S/A is V _MThe time, S/A changes very sensitive to input.Row decoder 12 is selected a word line based on Input Address 21.When RY 6 was high, row decoder 12 and bit line all lost efficacy idle circuit piece 18 ', and promptly all word lines and bit line be preliminary filling/be discharged to V all _M Bias circuit block 16 produces SE5 by a timing signal TS 8T.When reading to begin, SE 5 is low, and all data S/A do not work.After TS 8T uprised, SE 5 was put height, and all data S/A take a sample.This sampling process is till all outputs 8 all become effective output.RY 6 is sent then, finishes a read cycle.The sequential relationship of reading of 3D-ROM is described by Fig. 5.

In most of read cycle, the voltage increasing degree on the bit line is not enough to trigger its S/A.If at this moment all S/A are in running order, then they can consume a large amount of electric energy, but their output is invalid.It is in running order preferably at this moment only to stay a small amount of S/A, and they detect the change in voltage on its bit line.Have only when they find that this change in voltage is enough big, other S/A just is opened and takes a sample.Correspondingly, most of S/A only works in the time in the sub-fraction of read cycle, and this can cut down the consumption of energy.A purpose of self-timing that Here it is.

Fig. 2 C represents a kind of implementation method of self-timing.In 3D-M array 0A, increase by one first timing bit line 30T.It is preferably the bit line of location decoder 12 of where leaving one's post farthest, between it and every word line (20a...) that intersects with it diode connection (1aT...) is arranged simultaneously.In read procedure, the change in voltage speed on it cans be compared to the slowest data " 1 " bit line most and also wants slow.Like this, when the voltage on the first timing bit line 30T can trigger its S/A17T, the voltage on all data " 1 " bit line all should be even as big as triggering the degree of its S/A 17a-17d.At this moment just opening data S/A 17a-17d takes a sample.

Fig. 2 C has also described S/A piece 18, bias circuit block 16, row decoder 12 and bit line and has made idle circuit piece 18 '.

S/A piece 18 contains regularly S/A 17T of a plurality of data S/A 17a-17d and first, and they amplify the signal on the data bit line and the first timing bit line respectively.The change in voltage of S/A (17a...) on the bit line that is attached thereto surpasses its threshold voltage V _TThe time, output switching activity.Here, data S/A 17a-17d is controlled by SE5, and they are sampling when SE 5 is high only, can cut down the consumption of energy like this.The first timing S/A17T monitors the voltage on its bit line 30T when reading always, so its offset signal 5T is constant in read procedure.

Bias circuit block 16 is according to the size of the output 8T decision SE5 of the first timing S/A 17T.It contains a timing circuit 15T and an offset generating circuit 15B.Timing circuit 15T control timing signal 5d, offset generating circuit 15B produces corresponding offset signal SE5 and 5T.When 5d puts when low, SE5 puts height.

Row decoder 12 contains a standard row decoder 11C and a plurality of row decoder makes idle circuit piece 11a-11d.As RY6 when being high, decoder 12 lost efficacy, all word lines all with V _M7 short circuits.When RY 6 be low and 20a ' when high, word line and V _RJoin, promptly enter read states.

Bit line make idle circuit piece 18 ' by switch (being transistor 17a '-17d ') with all bit lines and V _M7 short circuits, the control end of these switches 17a '-17d ' all link to each other with same signal RY 6.As RY 6 when being high, all bit lines all with V _M7 short circuits.Bit line makes idle circuit piece 18, and the full-time course pattern is achieved.

Below describe a kind of flow process of reading of 3D-M nuclear 0 among Fig. 2 B and Fig. 2 C, its sequential chart is seen Fig. 5.Preferably use the full-time course pattern when reading, promptly when once reading, the data in all storage elements on the word line are read simultaneously.3D-M nuclear 0 is in default conditions at first, and promptly all word lines and bit line all are offset to V _M, and all data S/A all do not take a sample.At the rising edge of RD 4, register 12l captures AS2 (as " 00 ") and is sent to row decoder 12, then with the corresponding word line 20a in this address on voltage be raised to and read voltage V _RAnd the bit line (30a..) that has diode to link to each bar and it charges.At this moment, all data S/A17a-17d all do not take a sample, but the first timing S/A17T is monitoring the voltage on its bit line 30T always.When this change in voltage surpasses V _TThe time, output 8T uprises.Correspondingly, SE5 is put height, and all data S/A 17a-17d begin the bit-line voltage that they link to each other is separately taken a sample.After producing output DO 8, SE5 puts low, and data S/A 17a-17d is disconnected; Word line 20a also there is no need to remain on V again simultaneously _RCorrespondingly, RY 6 puts height, and 3D-ROM nuclear 0 is got back to default conditions.Finish a read cycle T like this.

Fig. 3 A-Fig. 3 G describes the various circuit that multiple 3D-M nuclear uses.Fig. 3 A-Fig. 3 CC describes multiple difference S/A.Fig. 3 DA-Fig. 3 DD describes second timing bit line and a kind of timing circuit 15T thereof.Fig. 3 E-Fig. 3 G describes a kind of offset generating circuit 15B, row decoding makes idle circuit 11a and V _MProduce circuit 14.

Because S/A requires to have extremely strong anti-interference, preferably can use difference S/A.Difference S/A removes one and is input as by the voltage of sense bit line, and it also needs a reference voltage.This reference voltage can provide by a umbral position line.Fig. 3 A represents two by being connected between sense bit line (30a, 30z), a umbral position line 30D and they and the difference S/A (17a, 17z).Umbral position line 30D can be shared by a plurality of S/A, with its every word line place of intersecting a diode 1aD is being arranged all.In read operation, the voltage on the umbral position line 30D is preferably between data " 1 " bit-line voltage and data " 0 " bit-line voltage.

Fig. 3 BA is the circuit diagram of first difference S/A nuclear 17C.It uses NMOS to 51a, 51b as input transistors, and the PMOS of mirror image symmetry to 51d, 51e as load.Its supply voltage is V _S/AAnd GND.Notice V _S/AMay be different from chip power voltage.Offset signal B controls tail current by NMOS 51c.Fig. 3 BB represents that one uses the data S/A of first difference S/A nuclear 17C.It also contains a latch 17L who is made of NMOS 51g and inverter 51h.Under the control of latch signal 5 ', NMOS 51g closes when SE5 uprises, but opens during prior to SE 5 step-downs.Like this, even S/A nuclear 17C does not take a sample, output 8a still remains unchanged.Fig. 3 BC represents that one uses the first timing S/A of first difference S/A nuclear 17C.This timing S/A takes a sample all the time.Inverter 51i, 51i form a latch 17TL, and 51l, 51m adjust waveform.When each read cycle began, NMOS 51k was under RD4 control, with latch 17TL zero clearing (equilibrating).

Fig. 3 CA is the circuit diagram of second difference S/A nuclear 17C '.Compare with Fig. 3 BA, it use cross-linked PMOS to 52d, 52e as load.Offset signal B controls tail current by NMOS 52c.When B is low, the output o+ of S/A nuclear, o-keeps its level before the B step-down, so this S/A nuclear is a latch.Fig. 3 CB represents the data S/A that another uses second difference S/A nuclear 17C '.Inverter 52f is used for adjusting waveform.Fig. 3 CC represents the first timing S/A that another uses second difference S/A nuclear 17C '.This timing S/A takes a sample all the time.When each read cycle began, NMOS 52g was under RD4 control, with the 17C ' zero clearing (equilibrating) of difference S/A nuclear.

Fig. 3 DA-Fig. 3 DD represents the design of multiple timing circuit 15T.Timing circuit 15T combines with offset generating circuit 15B, bias voltage SE 5 that can control data S/A.After 8T uprised, it was raised SE 5 and allows all data S/A begin sampling; Through one section delay, after promptly all data S/A were all effectively exported, it cut off SE 5, and then the sampling of end data S/A.In order to realize this delay, the embodiment of Fig. 3 DA has increased by one second timing bit line 30T ', the size of its S/A17T ' control lag in the 3D-M array.Here, second regularly bit line 30T ' with its crossing word line place one diode 1aT ' is arranged all at all, but the more general data S/A of its S/A17T ' is slow.When its output 8T ' when upset, the output of all data S/A should be ready, so sampling that can end data S/A.Clearly, this also can cut down the consumption of energy.Notice the first regularly beginning of bit line 30T control data S/A sampling, the second regularly end of bit line 30T ' control data S/A sampling.Fig. 3 DB represents the circuit diagram of the S/A17T ' that a kind of second timing bit line 30T ' uses.Compare its output with general data S/A (Fig. 3 BA).One unnecessary load capacitance 51C can be arranged, and also can be that the raceway groove of its input or load transistor 51a ', 51b ', 51d ', 51e ' is longer, or the like.Like this, the data S/A that this S/A17T ' is more general is slow.

Fig. 3 DC is the circuit diagram of a kind of timing circuit 15T.The output 8T ' of the second timing bit line 30T ' can directly be used as RY 6, and it combines with the output 8T of the first timing bit line 30T, produces bias control signal 5d, and 5d is by offset generating circuit 15B control SE5 (referring to Fig. 3 E) then.Fig. 3 DD is the circuit diagram of another kind of timing circuit 15T.Compare with Fig. 3 DC, a state control signal 6E who provides for external circuitry (as the circuit in 3DiM but outside 3D-M) is provided for it.As 6E when being high, 3D-M enters default conditions (all word lines and bit line meet VM), can not carry out any operation.At this moment, 3D-M is in " soft outage " state.When " soft outage ", 3D-M is consumed power not; In case but 6E put low after, 3D-M can enter operating state fast.Compare with " hard outage " (i.e. all word lines and bit line ground connection), 3D-M can " revive " quickly, and the speed of promptly resuming work is faster.This design can be used in the multiple application, as word line redundant circuit and motor-driven decoding circuit (is that the defective bit line maybe needs to be upgraded when replacing by readout word line at this 3D-M), or based on the integrated circuit testing of three-dimensional storage (when the tested circuit operate as normal).

Fig. 3 E is an offset generating circuit 15B.Current source 53a can be current source or sheet extrinsic current source in the sheet.Bias voltage 5T is produced by a NMOS53b who takes diode to connect.When 5d when low, 5T is transferred into SE5.As 5d when being high, SE5 ground connection.

Fig. 3 F is that a row decoder makes idle circuit 11a.When RY 6 was high, NMOS54b was switched on, word line 20a and V _M7 short circuits.When 20a ' height and RY6 when low, PMOS 54c is switched on, word line 20a with read voltage V _RShort circuit.Notice V _RThe supply voltage V that may be different from chip _Dd(referring to Fig. 8 CA).

Fig. 3 G is a V _MProduce circuit 14.It uses the S/A nuclear 55a identical with S/A, and contains a pressurizer (containing transport and placing device 55b and driving N MOS55c).All input and output of S/A nuclear are by short circuit, thus generation turnover voltage V _M7 '.In general, V _M～V _S/A/ 2.Pressurizer makes V _MThe output that produces circuit 14 remains on V _M, and enough big electric current is provided, so V _M7 is constant voltage dc sources.

Fig. 4 AA-Fig. 4 AD describes the temporal characteristics of bit-line voltage.As described in Fig. 4 AA, the voltage on word line 20y is risen to V _RAfter, word line 20y begins to charge by diode 1yj pairs of bit line 30j.Voltage on the bit line 30j is from its initial value (V _M) raise, the speed of its rising is decided by the speed of the parasitic capacitance charging of diode current pairs of bit line 30j.In general, bit line parasitic capacitance 1jC comprises: the junction capacitance 1j2 (corresponding to " 1 " storage element) of the coupling capacitance 1j0 of word line 20x and bit line 30j (corresponding to " 0 " storage element), back-biased diode 1zj, and adjacent bit lines 30i and 30k between coupling capacitance 1j3 and 1j4, and other interconnection line layer between coupling capacitance 1j1.Voltage on the bit line 30j is higher than V in read procedure _M, and except that all being in V by other word line 20x, 20z the readout word line 20y _MSo, have leakage current to flow on other word line 20z from bit line 30j by diode 1zj.The discharge effect that this leakage current pairs of bit line 30j produces is opposite with the charging effect that word line 20y produces.

Fig. 4 AB one is used for simulating the equivalent electric circuit of bit-line voltage temporal characteristics.Change in voltage Δ V on the word line 30j _bDetermine by three factors: diode 1yj, parasitic capacitance 1jC and equivalent diode 1jD.Equivalent diode 1jD is composed in parallel by n diode.Here, n is that all link to each other with bit line 30j and are in the number of back-biased diode.Under the worst reading mode, n equals N _WL-1.When the forward current of diode 1yj equals the reverse current of equivalent diode 1jD, Δ V _bReach static balancing voltage Δ V _Be

Fig. 4 AC represents current/voltage (IV) characteristic of diode 1yj.Its forward current I _f(V) 1f is greater than its reverse current I _r(V) 1r.Can find static balancing voltage Δ V under the worst reading mode with image method _BeAt first reverse IV curve be multiply by N _WL-1, V then moves right it _R-V _MCurve 1rs that obtains like this and the intersection point of 1f are the static balancing electromotive force Δ V under the worst reading mode _BeWrite as the form of equation,

I _f(V _R-V _M-ΔV _be)＝(N _WL-1)×I _r(ΔV _be)≈N _WL×I _r(ΔV _be) eq.(1)

Fig. 4 AD is the sequential chart of this bit line potentials.Bit line potentials appreciation Δ V _bFinally reach its static balancing voltage Δ V _BeAt moment τ, Δ V _bSurpass VT, the output of S/A becomes effective output, so τ is for visiting the time first.Pairs of bit line 30j,

τ _30j～V _T×C _30j/I _f eq。(2)

According to Fig. 2 C and Fig. 3 A, the temporal characteristics of the first timing bit line and umbral position line is different with data bit line.Correspondingly, their design is best different with data bit line.Fig. 4 BA-Fig. 4 CC explains and provides several designs.Fig. 4 BA represents a data bit line 30a and a reference bit lines 30r.Reference bit lines 30r can be one first timing bit line or umbral position line.In read procedure, the change in voltage Δ V on the reference bit lines 30r _30rPreferably should be slower than the change in voltage Δ V on data " 1 " the bit line 30a _30aFor umbral position line, best Δ V _30r～Δ V _30a/ 2 (Fig. 4 BB).According to eq. (2), can reach this purpose by the parasitic capacitance 1rC that increases on the reference bit lines 30r.Fig. 4 CA-Fig. 4 CC has represented several implementation methods.

Fig. 4 CA represents the first reference bit lines 30r.It is wideer than data bit line 30a, so it has bigger parasitic capacitance.Fig. 4 CB represents the second reference bit lines 30r.It comprises two continuous inferior bit line 30r1 and 30r2.They and general data bit line 30a have same widths.Inferior bit line 30r1 has diode to be connected with every word line that intersects with it, and inferior bit line 30r2 does not have diode to be connected with any word line.Correspondingly, the parasitic capacitance on the reference bit lines 30r is bigger, and the voltage climbing speed is slower.Notice that the length of inferior bit line 30r2 can be regulated by layout design.Fig. 4 CC represents the 3rd reference bit lines 30r, and it links to each other with a physics electric capacity 1r0.Physics electric capacity 1r0 can be mos capacitance (input capacitance that comprises S/A), metal capacitance or other conventional electric capacity.These electric capacity also can postpone to visit first time τ.

Fig. 4 D is illustrated in data bit line in the 3D-M array, umbral position line and the regularly design of bit line.Two set of bit lines D1, D2 are arranged in this embodiment, the shared umbral position line 30D of all data bit lines in each set of bit lines.This umbral position line 30D contains bit line 30D1,3,0D2 two times.In the 3D-M array, also has the first bit line 30T and regularly as the mute timing bit line 30TD of its reference bit lines.The first timing bit line 30T contains bit line 30T1,3,0T2 two times, and mute regularly bit line 30TD contains bit line 30TD1-30TD4 4 times.This embodiment also contains the second timing bit line 30T ', and it only contains bit line one time, but its S/A17T ' is slower.Clearly, the change in voltage on the umbral position line 30D and the first timing bit line 30T is slower than data bit line 30a, and the change in voltage on the mute regularly bit line 30TD is slower.

In fact, the umbral position line 30D and the first timing bit line 30T can adopt some simple designs.Because umbral position line 30D need drive lot of data S/A, the input capacitance of these data S/A makes the change in voltage on the umbral position line slack-off a lot, so umbral position line 30D can only use bit line one time.On the other hand, the first timing bit line 30T also can only contain bit line one time.At this moment, its S/A is preferably slower, but should be faster than the S/A of the second timing bit line 30T '.

Fig. 5 is the sequential chart of various signals in the 3D-M nuclear 0.At moment τ _30a, the voltage on the data bit line 30a has surpassed the V of its S/A17a _TBut because at this moment S/A does not open, so there is no valid data output.At moment t1, the first timing bit line 30T triggers its S/A17T, and this represents that all data S/A can begin sampling.At this moment, SE5 is sent, and all data S/A start working.At moment τ, when the second timing bit line 30T ' triggered its S/A17T ', this represented that all data S/A all finish sampling.All data S/A are disconnected.Finish a read cycle like this.

Eq. (2) and Fig. 4 AA provide a kind of design guideline for 3D-M.In order to shorten the time of visiting first, preferably can reduce the parasitic capacitance 1jC of bit line.Because a big chunk of bit line parasitic capacitance comes from coupling capacitance 1j3,1j4 between the wall of bit line limit, 3D-ROM preferably can use thin bit line.Though thin bit line has bigger series resistance, because the main resistance that the time is visited in decision first is from the 3D-ROM film, so the resistance increase of using thin bit line to cause can't have too many influence to visiting the time first.In addition, when the full-time course pattern, word line will provide read current for all bit lines, and its electric current is generally bigger.In order to reduce parasitic voltage drops and to overcome electromigration problems such as (electro-migration), 3D-M preferably can use thicker word line.Fig. 2 A represents the 3D-ROM structure of thicker word line of a kind of use and thin bit line.

B. with the 3D-M (3DcM) of data buffer zone

The performance of single 3D-M unit still is difficult to compare with conventional storage element.By the system integration, the potential of 3D-M could be developed fully.From collective's performance, 3D-M can compare with conventional memory, even better.Cached 3D-M (abbreviating 3DcM as) is the exemplary of 3DiM.It contains the integrated with it embedded RAM (eRAM) of a 3D-M and, and improves its reading speed by the time of visiting first of hiding 3D-M.Concerning external circuit, 3DcM can be regarded as a single memory: eRAM and be formed in the substrate, and 3D-M is stacked on the eRAM, and has kept a backup of 3D-M data among the eRAM.When data were sought by system from 3DcM, it was sought from eRAM earlier, as " hitting ", then directly reads from eRAM; As " miss ", then will be from 3D-M read data, and a backup left among the eRAM.Correspondingly, eRAM is the cache of 3D-M.As " hitting ", the time of visiting first of 3DcM is exactly the time of visiting first of eRAM, and the time of visiting first of 3D-M can't be discovered by the external world; As " miss ", 3DcM visit the time first and 3D-M is close.If the eRAM capacity is enough big, the probability that hits is bigger, can reduce the time of on average visiting first like this.The bandwidth of 3DcM is generally controlled by eRAM.

The class of operation of the read operation of 3DcM and computer high speed buffer storage seemingly.Fig. 6 A-Fig. 6 G is to the details of 3DcM, and especially interior data stream has been done more detailed description.Fig. 6 A represents the I/O port of a kind of 3DcM 0C, and it comprises that Input Address AS 73,3DcM read enabling signal cRD 75,3DcM ready for data signal cRY 77, clock signal C K 71 and data output DO 79.

Fig. 6 B is the block diagram of a kind of 3DcM 0C.It contains 3D-M nuclear 0, column decoder 70, eRAM 72, control circuit piece 74 and reads and select piece 76.In this special case, the size of 3D-M nuclear 0 is 1024 * 1024.When read operation, from the 3D-M array, select a page data (1024) and it is delivered to output 8 according to row address 2 (AS73 preceding 10 [13:4]).Here, one page 3D-M data are meant in the 3D-M array all data on the word line.Column decoder 70 is chosen a word (64) according to column address 2c (end of AS73 4 [3: 0]) again from this output page (1024).The word of choosing and with it corresponding address be copied among the eRAM 72.Data between the control circuit piece 74 control 3D-M nuclears 0 to eRAM 72 flow.Person skilled in the art scholar can design control circuit piece 74 according to the data flow of Fig. 6 D at an easy rate.Reading and selecting piece 76 decision dateouts 79 is from column decoder 70 or from eRAM 72.

Fig. 6 C represents a kind of eRAM 72.It contains a read-write Enable Pin R/W 74r and a hit/miss output H/M 72h.It also contains an eRAM data block 72D and an eRAM tag block 72T.ERAM data block 72D stores the 3D-M data, and every row of eRAM tag block 72T is stored the address tag of data in the eRAM data line corresponding with it.In this embodiment, the size of eRAM data block 72D is 64 x 64, and the size of eRAM tag block 72T is 8 x 64.AS 73[13:6] preceding 8 2a be stored among the eRAM tag block 72T AS 73[5:0] back 6 column address 2b that are used as eRAM 72.This eRAM 72 also has a comparator 72C.When read operation, it compares address tag 72to among the eRAM tag block 72T and address 2a.If they are identical, promptly hit, then export 72h for high; Otherwise 72h is low.

Fig. 6 D describes a kind of read operation of 3DcM.At first, after receiving cRD 75, AS73 is sent to eRAM 72, and eRAM 72 enters reading mode (step 91).Then, 72h carries out different operating (step 92) according to the H/M signal: as hit, then the data 79a that directly will read from eRAM 72 delivers to output 79 (steps 97), and sends cRY 79 (step 98); As miss, then need read data from 3D-M nuclear 0.This comprises following steps: at first, send RD4 signal (step 93); From 3D-M, read a page data then, and send RY6 (step 94); At this moment, eRAM 72 enters WriteMode, and word 79a and address 2b thereof that column decoder 70 is chosen are deposited in eRAM 72 (step 95); At last, data 79a or 79b are delivered to output 79 (steps 96), send cRY 79 (step 98) again.

When step 96, data can directly be read at the column decoder place when the 3D-M data are sent to eRAM 72.The time of visiting first of this method will lack.Fig. 6 EA represents a kind of corresponding data selector 76.It uses a MUX 76M.According to the size of its control signal 79s (generally by H/M signal 72h decision), MUX 76M decision output 79 is adopted from the data 79a (miss situation) of column decoder 70 or from the data 79b (hitting situation) of eRAM 72.

In addition, even miss, also can be after the 3D-M data copy to eRAM from eRAM 72 sense data.This method more easily satisfies the requirement of redundant circuit and software upgrading.Fig. 6 EB represents a kind of flow process of reading, and this reads the part that flow process is a step 96 among Fig. 6 D.After the 3D-M data are downloaded to eRAM 72, repeat the read operation ( step 91,92,97 etc. that comprises Fig. 6 D) of eRAM.Particularly, after step 95 was finished, AS 73 was delivered to eRAM 72 once more, and reading (step 96a).Because be certain to " hitting " specifically, promptly H/M signal 72h is high (step 96b) certainly, and the sense data 79b of eRAM is directly delivered to output 79 (step 96c).Fig. 6 EC represents a kind of data selector 76 that this method is used.Because of all dateouts all from eRAM 72, this data selector is a simple transmission gate 76T, its decision whether eRAM 72 data 79b are exported.

The embodiment of Fig. 6 B-Fig. 6 EC promptly may have only a word (64) to be copied to (other word may all be wasted) among the eRAM 72 based on " word duplicates " in the output page (1024).In order to make full use of the data of reading at every turn, preferably use " page or leaf duplicates ", promptly all words on the output page all are copied among the eRAM 72." page or leaf duplicates " can improve and read efficient.Fig. 6 F represents the 3DcM of a kind of use " page or leaf duplicates ".Different with Fig. 6 B is, its column address 2c ' is not 4 at the end of AS 73, but produces by control circuit piece 74 ' is inner.For being familiar with this professional personage, can be at an easy rate go out control circuit piece 74 ' according to the flow scheme design of reading of Fig. 6 D and Fig. 6 G.Fig. 6 G represents that a kind of column address produces flow process, and it is the part of step 95 among Fig. 6 D.After step 94, under 74 ' control, produce the address (step 95a) of all words on the output page successively, then, selected word and address thereof are copied to eRAM 72 (step 95b).Repeating step 95a, 95b reach its preset maximum value (step 95c) up to 2c '.Like this, output page all is copied among the eRAM 72.The eRAM 72 that uses during Fig. 6 H represents a kind of " page or leaf duplicates ".In this embodiment, the size of eRAM data block 72D still is 64 * 64, but it is divided into 4 eRAM sectors.The size of each eRAM sector is 64 * 16, and stores all data (1024) in the output page.Each sector uses an address tag capable.Correspondingly, the size of eRAM tag block 72T can be 8 * 4.

C. program speed

The user of 3D-EPROM can programme.In order to shorten the chip programming time, best a plurality of storage elements can be programmed simultaneously.This promptly is the notion of parallel programmed.Fig. 7 A represents a kind of implementation method of parallel programmed.In this special case, 3D-EPROM 1cb of unit and 1cc are programmed simultaneously.When programming, the voltage on the word line 20c rises to V _Pp, the voltage on bit line 30b, the 30c reduces to 0, and the voltage of all other address selection lines is V _Pp/ 2.Therefore, the voltage that is added on storage element 1cb, the 1cc is V _PpSo they are programmed simultaneously.For the voltage at least two bit lines is reduced to 0, column decoder is parallel columns decoding (Fig. 7 B) preferably.It has used two inferior decoder 70a, 70b.These inferior decoder 70a, 70b have identical column address 2C.They can be adjacent, also can be cross one another.In this embodiment, they are mirror image symmetries.Column address 2C (as " 1 ") is delivered among this Asia decoder 70a, the 70b simultaneously, and this reduces to 0 with the voltage on bit line 30b, the 30c, thereby can satisfy the voltage request of Fig. 7 A.

In order to reduce the number of encapsulation pin, United States Patent (USP) 6,385, V in 074 suggestion use a slice _PpGenerator.V in this sheet _PpGenerator utilizes chip power voltage V _DdProduce V _PpThis design is necessary for the frequent 3D-M that need programme.But for the 3D-M of " disposable " programming, they do not need frequent programming; Especially for the 3D-EPROM (as the PonC among Fig. 3) as data carrier, they are generally in factory's the inside programming (as by the data publisher).In use, user-readable, and do not programme.Concerning these are used, V _PpGenerator there is no need, and the chip area of its saving can be used for designing other function.Fig. 7 C describes a kind of V of having _PpThe 3D-M of wiring pad 12P, 70P.These wiring pads provide extraneous program voltage.For factory program, as the 3D-EPROM of data carrier, its programming generally is the chip level programming, so these wiring pads do not need to link to each other with package lead.This can reduce the number of encapsulation pin.

The 3D-EPROM of factory program can adopt the business model of internet, promptly utilizes the internet to come the desired data that write chip of transmission user.Simultaneously, factory (as the data publisher) can also have a plurality of databases, and these databases have a plurality of files.The user only needs to click the pointer (pointer) of required file on factory's webpage, factory just can propose required file from database, and write among the 3D-EPROM (referring to Fig. 8 AA and Figure 39 A of the PCT application of submitting to by same inventor " low cost photo etching technique ", be easy meter, these figure that do not draw in this specification only need to change the photoetching programming system among these figure into electric programing system and get final product).

2. the capacity of unit array

Shown in Fig. 8 AA-Fig. 8 AB, but the capacity of unit array has very big influence to the integration of 3D-M.For big unit array, the 3D-M chip can only contain a spot of array 0A of unit (Fig. 8 AA); For little unit array, the 3D-M chip need contain the array 0Aa-0Ai of unit (Fig. 8 AB) of larger amt.Because the peripheral circuit of unit array is positioned in the substrate, the unit array of larger amt means that substrate is by seriously fragmented.This fragmented substrate can make the layout design of substrate integrated circuit be subjected to very big restriction.In addition, the unit array of greater number can make the array efficiency step-down.But, preferably can use to have jumbo unit array in order to improve the integration of 3D-M.

The capacity C of 3D-M unit's array _AEqual its word line number N _WLWith number of bit N _BLProduct (Fig. 2 B, Fig. 8 B), therefore can be by improving N respectively _WLAnd N _BLImprove C _AFrom design point of view, N _BLGenerally unrestricted, therefore can adopt the unit array of rectangle.On the other hand, according to eq. (1) and make Δ V _Be=nV _T(n～2, V in general, _T～0.1V), N _WLCan be expressed as follows,

N _WL＝I _f(V _f)/I _r(V _r)＝I _f(V _R-V _M-nV _T)/I _r(nV _T) eq.(3)

N _WLBe subject to the positive and negative current ratio γ of storage element when reading.Here, the definition of γ is different with conventional definition: its forward bias voltage V _f(as～3V) can be much larger than reverse bias voltage V _r(as～0.3V).This benefits from the application of S/A and full-time course pattern etc.Eq. (3) are extremely valuable to the design of 3D-ROM.Clearly, can be by increasing V _RImprove N _WLOn the other hand, can improve N by using two polarization units _WLSo-called two polarization units are meant that electric current and the impedance that electric current ran on the rightabout flow through on its direction are far from it.

Fig. 8 B represents a kind of rectangle 3D-M array.In this embodiment, N _BLN _WLIn this chip, can place several this 3D-M arrays along the y direction.Like this, the final form of chip can roughly keep square.

Fig. 8 CA describes a kind of big V that utilizes _RImprove N _WLMethod.Here, V _RGreater than supply voltage V _DdBecause the IV characteristic of 3D-ROM film generally is an exponential type, so its read current I ₁(V _RThe time) be far longer than V _DdThe time electric current I ₂Therefore, N _WLAnd C _ACan increase a lot.Fig. 8 CB, Fig. 8 CC represent a kind of V _RProduction method.Fig. 8 CB is its circuit block diagram.V _RGenerator 12R is that voltage V is read in row decoder 12 generations _RIts general employing charge pump designs such as (charge-pump).Fig. 8 CC is a kind of V of containing _RThe substrate layout design of generator 12R.Here, the three-dimensional integrated V that makes _R Generator 12R can be arranged in substrate 0s, especially can be positioned at 3D-M array 0A below.

Except using big V _RCan also use two polarization units to improve C outward, _ATwo polarized films and/or two polarization structures can be contained in two polarization units.Two polarized films are that the difference by material produces two polarity effects (Fig. 8 D-Fig. 8 EC); Two polarization structures produce two polarity effects (Fig. 8 F-Fig. 8 GC) by the difference at interface.

Fig. 8 D explains the notion of two polarized films.Two polarized films 38 contain at least two level film 38a, 38b.Their material preferably has bigger difference.When electric current flows through two polarized films 38 (promptly from port 39a to port 39b) along direction 37a, it at first runs into time film 38a, runs into time film 38b then; On the other hand, when it flowed along direction 37b (promptly from port 39b to port 39a), it ran into time film 38b earlier, and then ran into time film 38a.This order that runs into time film 38a, 38b can greatly have influence on the size of electric current.A very familiar example is the p-n junction diode, and it causes the generation of diode phenomenon by using different doping types.Two polarized films 38 walk still farther than diode: except doping type, their sill can also be different.Here, the sill of a skim is the main material that constitutes this tunic.Fig. 8 EA-Fig. 8 EC has represented the embodiment of several two polarized films.

Fig. 8 EA represents first kind of two polarization 3D-ROM film.It contains two level film 32a, 32b, and they use different sills respectively, and as follows the sill of film 32a is a silicon, and the sill of inferior film 32b is silicon carbon alloy (Si _yC _1-y, 0＜y＜1).Other semi-conducting material is as germanium, germanium-silicon alloy (Si _zGe _1-z, 0＜z＜1), diamond also can be used as sill.Here, high bandgap semiconductor material such as silicon carbon alloy, diamond (referring to the semi-conducting material of band gap greater than silicon) have certain advantage, because they have hot properties preferably.Except semi-conducting material, two polarized films 38 can comprise: the composite membrane of semi-conducting material and dielectric material (for example, inferior film 32a contains the semiconductor material, and inferior film 32b contains a dielectric material); Different dielectric materials (for example, inferior film 32a contains amorphous silicon, and inferior film 32b silicon nitride comprising); The sill of different structure (for example, inferior film 32a has non crystalline structure, and inferior film 32b has polycrystalline or microstructure, and this also has expression in Fig. 8 EB); Different electrode materials (for example, uses the metal with different work functions; Or the metal of different interfacial characteristicses is arranged with the 3D-ROM film; Or an electrode uses metal, the semi-conducting material that another electrode use is mixed).These methods can improve the positive and negative current ratio of 3D-ROM unit.

Fig. 8 EB represents the 3D-ROM film of second kind of two polarization.In this special case, between electrode 31 and 3D-ROM film 32a, one deck micro crystal material 32au is arranged.If only at an electrode interface (as the interface of electrode 31 and 3D-ROM film 32) the crystallite film is arranged, then two of 3D-ROM film polarization are reinforced, and can obtain a bigger positive and negative current ratio like this; On the other hand, micro crystal material can reduce metal-semiconductor contact resistance, at least one electrode interface (as the interface of electrode 31 and 3D-ROM film 32; And/or the interface of electrode 33 and 3D-ROM film 32) increases micro crystal material and can add the heavily conducting electric current, shorten the time of visiting first of 3D-ROM.

Fig. 8 EC represents the 3D-ROM film of the third two polarization.In this embodiment, 3D-ROM film 32 contains a p ⁺Film 32p, v film 32x and n ⁺Film 32n.V film 32x is that n is low-doped or plain, and these films are based on amorphous silicon.The deposition sequence of these films is 32n, 32x, 32p.This structure can obtain〉10A/cm ²Forward current and＜6 * 10 ^-5A/cm ²Reverse current.

Fig. 8 F explains the notion of two polarization structures.Interface between 3D-ROM film 32 and the top electrode 31 is top interface 32ti, and the interface between the hearth electrode 33 is low interface 32bi.In a kind of two polarization structures, the shape difference at these interfaces: what a interface has a most advanced and sophisticated 33t who strengthens, and another interface is more level and smooth.Correspondingly, electronics is transmitted in a direction and is strengthened, thereby improves positive inverse ratio.

Fig. 8 G is a kind of embodiment of two polarization structures.In this embodiment, hearth electrode 33 has polycrystalline structure, and its surperficial 32bi is more coarse; After 3D-ROM film 32 was deposited on the hearth electrode 33, non-crystalline material wherein made the interface 32ti between it and the top electrode 31 become comparatively level and smooth.Therefore, 31 electronics emission is enhanced from hearth electrode 33 to top electrode, and promptly flowing to the electrorheological of hearth electrode 33 from top electrode 31 must be big than the electric current on the rightabout.Therefore, can hearth electrode 33 be used as bit line with top electrode 31 as word line.

3. the raising of rate of finished products

Defective can cause various forms of read errors and reduce rate of finished products.Shown in Fig. 9 AA-Fig. 9 CB, the 3D-M array has six kinds of defectives, comprising: 1. open circuit 20o (Fig. 9 AA), 2. word line short circuit 20s (Fig. 9 AB), 3. open circuit 30o (Fig. 9 BA), 4. bitline short circuits 30s (Fig. 9 BB), 5. too small 3D-ROM unit forward current (Fig. 9 CA), 6. excessive 3D-ROM unit reverse current (Fig. 9 CB) of bit line of word line.

To word line defect 1 and 2, the whole piece word line can not be read correct data, and this causes word line failure.Pairs of bit line defective 3 and 4, whole bit line can not be read correct data, and this causes the bit line mistake.The forward current 1f ' of 3D-ROM defective unit 5 is too little, causes Δ V _BeCross lowly, S/A can not be triggered, thereby logical zero (Fig. 9 CA) is misread into by a logical one unit.Fortunately, this defective only can cause the individual bits mistake.The reverse leakage current of 3D-ROM defective unit 6 is too big, and when reading when being in other storage element on the same bit lines with this defective unit, it can limit Δ V _Be, S/A can not be triggered, thereby can not read valid data (Fig. 9 CB).This defective can cause the bit line mistake.Defective 5 and 6, especially 6, very big to the intrinsic rate of finished products influence of 3D-M array.

Be to improve rate of finished products, can use seamless 3D-ROM unit, it directly reduces the defective number (Figure 10 A-Figure 12 B) in the 3D-ROM array.In addition, also can use multiple 3D-M error correction scheme, as error correcting code (ECC), redundant circuit etc., error correction scheme is corrected in the 3D-M array mistake (Figure 13-Figure 15 C) that defectiveness causes.

A. seamless 3D-ROM unit

Defective in the 3D-ROM array may be introduced at the several stages of technological process, promptly before the 3D-ROM film forms (as to the hearth electrode top), in the 3D-ROM film forms (to the 3D-ROM film), after the 3D-ROM film forms (as to 3D-ROM film top).The cleannes of these films (being 3D-ROM film and adjacent at least with it part hearth electrode and part top electrode) are very big to the rate of finished products influence of 3D-ROM, so they are called as the defective sensitive membrane.A processing step common, that be easy to introduce defective is the figure conversion.In the figure transfer process, silicon chip will pass through photoetching and etching steps such as (or complanations).All these steps can be introduced extraneous objectionable impurities or damage 3D-ROM film.Therefore, in the forming process of defective sensitive membrane, should avoid the figure switch process.

In order to improve the intrinsic rate of finished products of 3D-M, the present invention proposes a seamless 3D-ROM unit.Figure 10 A represents a kind of seamless 3D-ROM unit.It contains hearth electrode 64,3D-ROM film 62, top buffer film 60 and top electrode 66.Top electrode 66 contains top buffer film 60 and top conductor 65, and they link to each other by access opening (opening) 67.Interface between top buffer film 60 and the 3D-ROM film 62 is top interface 62ti, and the interface between 3D-ROM film 62 and the hearth electrode 64 is bottom interface 62bi.In seamless 3D-ROM, at least a portion top buffer film 60 has identical cross section with at least a portion 3D-ROM film 62.At its technological process (Figure 11 AA-Figure 11 E,) in, 3D-ROM film and adjacent at least with it part hearth electrode and part top electrode are to form with a kind of seamless form: do not have the figure conversion between these processing steps, so can not introduce impurity to top interface 62ti and bottom interface 62bi.This technological process preferably can be carried out in a boundling equipment (clustertoo1).Figure 10 B represents another kind of seamless 3D-ROM unit.Its opening 67 has used nF opening mask in forming process, so its size is longer than the length of side of top buffer film 60.

The kinds of processes flow process of 3D-ROM unit employing that Figure 11 AA-Figure 11 E ' expression is seamless.In Figure 11 AA, all defect sensitive membrane comprises that hearth electrode 64,3D-ROM film 62 and a part of top electrode (promptly the top buffer film 60) are to form with a kind of seamless form.Therefore, the defective of top interface 62ti and bottom interface 62bi seldom.In addition, one deck etchant resist (etchstop layer) 60b (Figure 11 AB) can also be arranged between 3D-ROM film 62 and top buffer film 60, its function is described in Figure 11 BC.All these films (64,62,60b, 60) can form with seamless form.

Then, top buffer film 60 is carried out the figure conversion.Figure 11 BA-Figure 11 BC represents several 3D-ROM structures after this step.In Figure 11 BA, a part of hearth electrode 64 is exposed.In Figure 11 BB, a part of 3D-ROM film 62 is exposed.Figure 11 BC is the sectional view of structure after this step among Figure 11 AB.Etchant resist 60b can protect 3D-ROM film 62 in the place, and makes it not to be etched when this step.

After top buffer film 60 is shaped, preferably also want a step (Figure 11 CA-Figure 11 CC) of repairing 3D-ROM film 62 edges.This step is similar to the grid rear oxidation step (post-gate-oxidation) of conventional MOS technology.Figure 11 CA is the sectional view of structure after this step among Figure 11 BA, and a part of hearth electrode 64 converts medium 68d to by methods such as oxidations.Figure 11 CB is the sectional view of structure after this step among Figure 11 BB, and at least a portion 3D-ROM film 62 converts medium 68d to by methods such as oxidations.Figure 11 CC is the sectional view of structure after this step among Figure 11 BC, and at least a portion etchant resist 60b converts medium 68d to by methods such as oxidations.

Then, hearth electrode 64 is carried out the figure conversion, form 3D-ROM heap 69 (Figure 11 D).Medium 68 between the deposit low layer is opened access opening (opening) 67 then, and forms top conductor 65 (Figure 11 E).

Figure 11 D ', Figure 11 E ' are expressed as and form the required additional step of seamless 3D-ROM unit among Figure 10 B.After forming 3D-ROM heap 69, deposit low layer medium 68 is also done complanation to it.Afterwards, formation is provided with medium 23 (Figure 11 D ') on this structure.Low layer medium 68 and medium 23 is set preferably contains different medium be a silica as low layer medium 68, and it is silicon nitride that medium 23 is set.Then, to nF opening mask exposure.In being set, medium 23 forms opening 67 by an etching step then.During etching, can select its prescription that it is stopped on low layer medium 68.At last, fill conductor material to form top conductor 65 (Figure 11 E ').

Figure 12 A and Figure 12 B describe the two kinds of seamless 3D-EPROM of standard units.A part of film (as accurate conductive membrane 62a) in these accurate seamless 3D-EPROM units is to form with seamless form; Another part film (as anti-fuse film 62b) then is to form in a usual manner.In Figure 12 A, accurate conductive membrane 62a is between top buffer film 60 and hearth electrode 64, and it forms with a kind of seamless form; And anti-fuse film 62b is between passage stopple 63 and top electrode 65, and it forms with conventional method.In Figure 12 B, accurate conductive membrane 62a forms with a kind of seamless form; Anti-fuse film 62b is between top buffer film 60 and top electrode 65, and it forms with conventional method.In these two special cases, the defective of accurate conductive membrane is few.Notice that the position of accurate conductive membrane and anti-fuse film can exchange.

B. error correction scheme

For improving the rate of finished products of 3D-M, can also use error correction scheme, comprise error correcting code (ECC) and/or redundant circuit etc.Figure 13 represents the 3D-M of a band ECC.This 3D-M comprises that one has 3D-M nuclear 0, column decoder 70 and the ECC circuit 110 of ECC sign indicating number.In 3D-M nuclear 0,1024 valid data are arranged on every word line, they are divided into 16 64 word.For Hamming code, each word needs 7 check and correction positions, so the data bit on every word line has 16 * (64+7)=1136.When reading, this 1136 bit data 8 is by column decoder 70 back outputs 71 bit data 79a '.ECC circuit 110 converts this 71 bit data 79a ' to 64 valid data 79a.

On the other hand, redundant circuit can be corrected the individual bits mistake, word line failure and bit line mistake.Figure 14 A represents first kind of 3D-M with redundant circuit.It contains 3D-M nuclear 0, column decoder 70, alternative MUX (mux) 116S, 116B, 116W and three redundant circuit pieces of three groups 64.The redundant circuit piece comprises individual bits redundant circuit piece 118S, bit line redundant circuit piece 118B and word line redundant circuit piece 118W, and they correct individual bits mistake, bit line mistake, word line failure respectively.The address and the error correction data of each redundant circuit piece storage defect (as defective unit, defective bit line, defective word line).When Input Address conformed to a defective addresses, the error correction data corresponding with this defective addresses was sent to an input (117S, 117B, 117W) of MUX (being mux) (116S, 116B, 116W).Choose at mux under the control of end (115S, 115B, 115W), error correction data is with the 3D-M output 79a of correspondence " ' replace.Individual bits redundant circuit piece 118S, bit line redundant circuit piece 118B describe in Figure 14 B-Figure 14 DC; Word line redundant circuit piece 118W can be used in the software upgrading, their motor-driven code blocks that is otherwise known as, and its details is described in Figure 15 B-Figure 15 C.

Figure 14 B represents a kind of individual bits redundant circuit piece 118S.This embodiment contains two error correction groups, and they can correct two place defective units.Clearly, 118S can contain more error correction group.Each error correction group contains a plurality of registers, and they store after preceding 4 bs1 of column address, row address ws1 (10), the column address of significance bit vs1 (1) and defective unit 6 bs1 ' and error correction bit ds1 (1) respectively.Each memory choose end by "〉" expression.Significance bit is represented the validity of this error correction group, and only when it was high, error correction group address stored and error correction data were just effective.The vs1 register choose the end 122s can with V _DdConnect, also can be connected with other clock signal (as 74r).When reading, comparator 121a, 121c will import column address 2c, AS2 and compare with bs1, ws1 respectively, as conform to, and then read bs1 ', ds1.Decoder 121D puts height according to bs1 ' (6) with the control end 115S of a mux 116S.Simultaneously, ds1 is transferred into the input 117S of mux 116S, replaces corresponding dateout 79a under the control of 115S ".Here, when vs1 is low or Input Address and defective addresses when not being inconsistent, signal 122D puts low and decoder 121D was lost efficacy, and all 115S are low, and then mux 116S does not carry out any data and replaces.

Figure 14 C represents a kind of bit line redundant circuit piece 118B.This embodiment contains two error correction groups, and they can correct two place's defective bit lines.6 bb1 ' and error correction row db1 (1024) after preceding 4 bb1 of column address of each error correction group storage significance bit vb1 (1) and defective bit line, the column address.The error correction row contain the error correction data of all data on the defective bit line.When reading, column address 2c and bb1 compare, as conform to, and then read bb1 ', db1.Decoder 123D puts height according to bb1 ' with the control end 115B of a mux 116B.Simultaneously, from db1, select corresponding error correction bit and deliver to the input 117B of mux 116B, under the control of 115B, replace corresponding dateout 79a ' according to AS2.

Redundant circuit piece among Figure 14 B-Figure 14 C is based on " error correction when reading ".In addition, can utilize the characteristics of a backup of the eRAM storage 3D-M data among the 3DiM, realize " reading the back error correction ", promptly the data among the 3D-M (comprising correct data and misdata) are downloaded to eRAM earlier, correct in eRAM then.Figure 14 DA has described a kind of redundant circuit piece 118SB based on " reading the back error correction ", and it corrects the individual bits mistake earlier, corrects the bit line mistake again.It contains individual bits error correction block 120S and bit line error correction block 120B, and they correct individual bits mistake, bit line mistake respectively.

Individual bits error correction block 120S contains the first error correction memory block 126S.This error correction memory block 126S contains a plurality of error correction groups, column address bs (10), row address ws (10) and the error correction bit ds (1) of each error correction group storage significance bit 126d (1) and defective unit.In this embodiment, effectively the error correction group is deposited from the bottom of 126S successively.After signal cRY79 puts height (being the eRAM data ready), 126S reads the error correction group line by line under the control of timing circuit 126a.Figure 14 DB represents a kind of circuit design of timing circuit 126a, and its function is: as long as significance bit 125d is for high, it can send counter clock signal 125a always; In case the 125d step-down is then sent counter O reset signal 125b and the individual bits error correction signal 79 ' that finishes.As long as so also have effective error correction group to be read out (125d is for high), the output 125c of counter 126b increases always, 125c is used as the address 125c of error correction memory block.Address decoder 126c reads an error correction group according to 125c.Comparator 126e is ws 125e and AS 2 relatively, as conforms to, and then bs125f is sent to the address end A[9:0 of eRAM 72], ds125g is sent to the data terminal D of eRAM 72, and will replace corresponding to the data of individual bits mistake among the eRAM 72.

Bit line error correction block 120B contains the second error correction memory block 126B.This error correction memory block 126B also contains a plurality of error correction groups.The column address bb (10) and the error correction row db (1024) of each error correction group storage significance bit 128d (1) and defective bit line.When receiving that the individual bits error correction finishes behind the signal 79 ', 128B begins to read the error correction group.128a uses the timing circuit identical with 126a.Similarly, as significance bit 127d when being high, counter 128b can increase the address 127c of 128B always.Address decoder 128c reads bb 127f according to 127c, and it is delivered to the address end A[9:0 of eRAM 72].128B selects required error correction bit 127g (1) according to AS 2 again from db, and it is delivered to the data terminal D of eRAM 72, thereby replaces the data corresponding to the bit line mistake.The sequential chart of above-mentioned " reading the back error correction " flow process is seen Figure 14 DC.

4. the upgradability of software

Software in use generally can experience repeatedly upgrading.In each escalation process, a part of source code (software code of initial issue) is upgraded sign indicating number and substitutes.It is generally acknowledged: come storing software as using masking film program read-only memory (MROM), after then chip dispatched from the factory, software can't be upgraded.To conventional MROM, this meets the fact.But to 3D-M, this viewpoint also is false.As previously mentioned, the 3D-M of storage source code can be at an easy rate integrates (being 3DiM) with the embedded RWM of routine, and these RWM can be used for storage upgrading sign indicating number, so the 3DiM support software is upgraded.Because upgrading sign indicating number take up space is more much smaller than source code, the capacity requirement of RWM is little, so the global storage cost is not high.

For the ease of software upgrading, the best modularization of software design.Figure 15 A represents the storage mode of a kind of software in 3D-M.Because the easiest data substitute mode is that word line is replaced among the 3D-M, the data that are about on the whole piece word line are replaced together, so software module is that unit leaves in the 3D-M array with the 3D-M page or leaf preferably, and had better not share same 3D-M page or leaf between the software module.Here, 3D-M page or leaf (as 20S[0]) is meant that a word line (as 20[0]) goes up all data of storage.In this embodiment, software module 160a contains 2047 bit data, because of a 3D-M page or leaf contains 1024 bit data, so 160b is stored in two 3D-M page or leaf 20S[0], 20S[1] in, wherein, 3D-M page or leaf 20S[1] last 1 1bz dummy argument preferably.If software module 160a need be upgraded, word line 20[0 then], 20[1] on all data when output, be upgraded sign indicating number and replace.This can realize by motor-driven code block.

Figure 15 B-Figure 15 C represents two kinds of motor-driven code blocks.These motor-driven code blocks also can be used to correct word line failure.The first motor-driven code block 166 among Figure 15 B is similar with Figure 14 B-Figure 14 C, and it is based on " replacing when reading ".This embodiment contains two upgrading groups, and they can be upgraded to two 3D-M pages or leaves.Each upgrading group storage significance bit vw1 (1) and the row address ww1 (10) and the upgrading sign indicating number dw1 (1024) that need the upgrading source code.The end 161s that chooses of significance bit register preferably links to each other with cRD75.When reading, comparator 162a is AS2 and ww1 relatively, as conforms to, and then reads 64 bit data 117W according to 2c from dw1, and under the control of word line replacement signal 115W dateout is replaced.Correspondingly, external circuitry is only seen the upgrading sign indicating number.In addition, motor-driven code block can not repeat them here based on " reading the back replaces " (referring to Figure 14 DA) yet.Notice, word line data is replaced, then in this read cycle, there is no need read data from the 3D-M array,, also can resume work rapidly thereby cut down the consumption of energy so the 3D-M array can be forced to enter " soft outage " (referring to Fig. 3 DD) as need.

The second motor-driven code block among Figure 15 C has been used the notion of paging management in the computer virtual memory, and it is considered as virtual address with Input Address, and it is carried out address transition and obtains the physical address of memory.This embodiment contains a upgrading piece 86O, an address decoder 164D and an address conversion block 164T.Upgrading piece 86O contains RWM, its storage upgrading sign indicating number.3D-M0 and upgrading piece 86O form a unified memory space 86S.Here, 3D-M0 has occupied the low 1020 row R[000000 0000 of storage and uniform space 86S]-R[11111 11011], upgrading piece 86O occupies high 4 row R[11111 11100]-R[11111 11111].Address conversion block 164T is actually a memory, address or the accurate address of its 86S of each row storage.So-called accurate address is meant that it need just can be regarded as physical address through after some computings.The Input Address 86A of address conversion block 164T is high 10 A[13:4 of Input Address], its output 86TA has 10 TA[9:0], they are finally delivered to the physical address of address decoder 164D as storage and uniform space 86S.Address decoder 164D provide address decoding according to physical address to 86S.When needs used source code among the 3D-M, physical address points to 3D-M0: as 86A was 0000000000 (being the capable 165a of 164T), and 86TA is 00,000 00000, the capable R[00000 00000 among its sensing 3D-M0], i.e. source code.When needs used the upgrading sign indicating number, physical address points to upgrading piece 86O: as 86A was 00,000 00100 (being the capable 165d of 164T), and 86TA is 11,111 11110, and its points to the capable R[11111 11110 of upgrading piece 86O], promptly upgrade yard.Address transition can be used in the application such as software upgrading, the error correction of defective word line, single core computer (computer-on-a-chip) easily.

Though above specification has specifically described examples more of the present invention, those skilled in the art should understand, under prerequisite not away from the spirit and scope of the present invention, can change form of the present invention and details, for example, 3D-M array implement example in this specification is 1024 * 1024, the actual 3D-M array that uses generally is～and 10 ⁴* 10 ⁴This does not hinder them to use spirit of the present invention.Therefore, except the spirit according to additional claims, the present invention should not be subjected to any restriction.

Claims (10)

1. three-dimensional storage, it is characterized in that containing: one contains substrate (Os) and a plurality of three-dimensional accumulation layer (100) that is stacked on this substrate top of a plurality of data sense amplifiers (17c...), each described three-dimensional accumulation layer contains at least one unit array (0A), this unit array contains many word lines (20a...), many data bit lines (30c...) and a plurality of storage element (1ac...) that is positioned at described word line and place, bit line crosspoint, described data bit line and the coupling of described data sense amplifier; In a read cycle, a word line is selected, and the voltage on this word line is for reading voltage (V _R), and all data on this word line were all read in this read cycle.

2. three-dimensional storage according to claim 1, its feature also is: this unit array also contains at least one the first bit lines (30T) regularly, this first timing bit line is parallel with described data bit line, also has a storage element at the place, crosspoint of this first timing bit line and described selected word line; In read procedure, the sampling of this data sense amplifier of signal enabling that produces on this first timing bit line.

3. three-dimensional storage according to claim 1, its feature also is: this unit array also contains at least one the second bit lines (30T ') regularly, this second timing bit line is parallel with described data bit line, also has a storage element at the place, crosspoint of this second timing bit line and described selected word line; In read procedure, the signal that produces on this second timing bit line stops the sampling of this data sense amplifier.

4. three-dimensional storage according to claim 1, its feature also is: this unit array also contains at least one umbral position line (30D), this umbral position line is parallel with described data bit line, also has a storage element at the place, crosspoint of this umbral position line and described selected word line; In read procedure, magnitude of voltage is between data " 1 " bit line and data " 0 " bit line on this umbral position line.

5. three-dimensional storage according to claim 1, its feature also is: this reads voltage greater than supply voltage (V _Dd).

6. three-dimensional storage according to claim 1, its feature also is: the number of bit (N of this unit array _BL) greater than word line number (N _WL).

7. three-dimensional storage according to claim 1, its feature also is: this three-dimensional storage is the masking film program three-dimensional storage.

8. three-dimensional storage according to claim 1, its feature also is: this three-dimensional storage is an electric programmed three-D memory.

9. three-dimensional storage according to claim 8, its feature also is: this electric programmed three-D memory is at least programmed three-D memory.

10. three-dimensional storage according to claim 8, its feature also is: this electric programmed three-D memory is repeatedly a programmed three-D memory.

Images