Vertical memory cell with non-self-aligned floating drain-source implant

Abstract

Various embodiments provide a memory cell that includes a vertical selection gate, a floating gate extending above the substrate, wherein the floating gate also extends above a portion of the vertical selection gate, over a non-zero overlap distance, the memory cell comprising a doped region implanted at the intersection of a vertical channel region extending opposite the selection gate and a horizontal channel region extending opposite the floating gate.

Claims

1. A method, comprising: etching a trench in a substrate; depositing a first dielectric layer on walls of the trench; forming a vertical selection gate extending in the trench; depositing on the substrate a second dielectric layer; forming a floating gate partially overlapping the vertical selection gate over a non-zero overlap distance; forming an electrically floating doped region at an intersection of a vertical channel region extending opposite the selection gate and a horizontal channel region extending opposite the floating gate.

2. The method of claim 1, wherein the forming of the electrically floating doped region includes implanting through a vertical wall of the trench.

3. The method of claim 2 wherein the implanting occurs before depositing the first dielectric layer on the walls of the trench.

4. The method of claim 1, further comprising forming an indentation that extends into the substrate along the vertical channel region.

5. The method of claim 1, further comprising forming a control gate on the floating gate.

6. The method of claim 1 wherein the forming the electrically floating doped region includes tilt implanting dopants.

7. A method, comprising: forming a trench in a substrate; forming a first memory cell on the substrate on a first side of the trench and a second memory cell on the substrate on a second side of the trench, forming each memory cell including: forming a selection gate in the trench; forming a floating gate on the substrate and on a portion of the selection gate; and forming a control gate on the floating gate; forming first and second doped regions in the substrate, the first and second doped regions being positioned on opposite sides of the selection gate; and forming first and second indentations respectively along walls of the selection gate.

8. The method of claim 7 wherein the forming of the first and second doped region includes tilt implanting dopants.

9. The method of claim 7 wherein the forming of the floating gate includes forming the floating gate in a respective one of the first and second indentations.

10. A method, comprising: forming a selection gate in a substrate; forming a first doped region along adjacent to a first side of the selection gate such that there is a first interface between the first doped region and the first side of the selection gate; forming a second doped region along adjacent to a second side of the selection gate such that there is a second interface between the second doped region and the second side of the selection gate; forming a first floating gate on the substrate on the first doped region and on a first portion of the selection gate, a sidewall of the first floating gate being positioned between the first interface and the second interface; and forming a second floating gate on the substrate on the second doped region and on a second portion of the selection gate, a sidewall of the second floating gate being positioned between the first interface and the second interface.

11. The method of claim 10 wherein forming the first and second floating gates include forming first and second extensions of the first and second floating gates, respectively, where the first extension is between the first doped region and the first side of the selection gate and the second extension is between the second doped region and the second side of the selection gate.

12. The method of claim 10, further comprising: forming a first control gate on the first floating gate; and forming a second control gate on the second floating gate.

13. The method of claim 10 wherein the forming of the first and second doped regions include tilt implanting dopants.

14. The method of claim 10, further comprising forming a third and a fourth doped region that are spaced from each other by the first and second floating gates.

15. A method, comprising: forming a selection gate in a substrate; forming a first doped region along a first side of the selection gate; forming a second doped region along a second side of the selection gate; forming a first floating gate on the substrate on the first doped region and on a first portion of the selection gate, forming the first floating gate including forming a first extension of the first floating gate, the first extension being between the first doped region and the first side of the selection gate; and forming a second floating gate on the substrate on the second doped region and on a second portion of the selection gate, forming the second floating gate including forming a second extension of the second floating gate, the second extension being between the second doped region and the second side of the selection gate.

16. The method of claim 15, further comprising: forming a first control gate on the first floating gate; and forming a second control gate on the second floating gate.

17. The method of claim 15 wherein the forming of the first and second doped regions include tilt implanting dopants.

18. The method of claim 15, further comprising forming a third and a fourth doped region that are spaced from each other by the first and second floating gates.

Description

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

(1) These objects and features of the present disclosure will be better understood in the light of the following description of embodiments of a method of manufacturing a memory cell according to the present disclosure, and of examples of memory cells produced according to this method, given in relation with, but not limited to, the following figures in which:

(2) FIG. 1 described above shows a conventional memory array architecture comprising split-gate memory cells,

(3) FIG. 2 described above is a cross-section of a conventional split-gate memory cell having a vertical selection gate,

(4) FIG. 3 described above shows a memory array architecture receiving the memory cell in FIG. 2,

(5) FIGS. 4 to 18 are cross-sections showing steps of a method of manufacturing a memory cell according to the present disclosure,

(6) FIGS. 19A to 19C show memory cells produced with this method,

(7) FIGS. 20 and 21 show complementary steps of manufacturing an integrated circuit comprising a memory cell according to the present disclosure,

(8) FIG. 22 shows a method of programming a memory cell according to the present disclosure,

(9) FIG. 23 shows a method of erasing a memory cell according to the present disclosure,

(10) FIGS. 24 and 25 show another method of erasing a memory cell according to the present disclosure,

(11) FIG. 26 shows a memory array architecture comprising a memory cell according to the present disclosure, and

(12) FIG. 27 shows an example of a memory circuit comprising memory cells according to the present disclosure.

DETAILED DESCRIPTION

(13) FIGS. 4 to 18 are cross-sections showing steps of a method of manufacturing a memory cell according to the present disclosure. FIGS. 13A and 13B show two alternatives of a step of this method. FIGS. 14A to 14C and 15A to 15C show three alternatives of two other steps of this method. FIGS. 19A to 19C show the three alternatives C31, C32, C33 of a memory cell C3 according to the present disclosure produced with this method and its alternatives.

(14) The memory cell C3 (C31, C32, C33) shown in FIGS. 19A, 19B, 19C comprises a horizontal floating gate FG formed on a P-type substrate PW, a horizontal control gate CG extending above the floating gate FG, the assembly forming a gate stack FG/CG, and a vertical selection gate SG formed in a trench 10 made in the substrate, the trench being covered with a dielectric layer D1. The memory cell C3 is here produced at the same time as a twin memory cell C3 (C31, C32, C33) using the same selection gate SG.

(15) The floating gate FG extends above a portion of the selection gate SG. The distance Dov between the proximal edge of the floating gate FG and the corresponding proximal edge of the selection gate SG is here negative, and is referred to as overlap distance below.

(16) The memory cell C3 also comprises an N-type embedded source region n0, linked to a source line SCL, an N-type floating drain-source region n1 which is not linked to any conductive line or source of potential, and an N-type drain region n2 intended to be linked to a bit line. The memory cell C3 can be seen as a combination of a horizontal floating-gate transistor and of a vertical selection transistor, the horizontal floating-gate transistor comprising the gate stack FG/CG, the region n2 as drain region and the region n1 as source region, the vertical selection transistor comprising the vertical selection gate SG, the region n1 as drain region, and the region n0 as source region. However, the drain-source region n1 is here floating and the assembly forms in reality the equivalent of only one transistor of the split-gate type, programmable by hot-electron injection and extending between the source region n0 and the drain region n2. Thus, the region n1 is referred to as drain/source region by analogy with a memory cell structure with two transistors of the above-mentioned type, but here does not collect electrons. The region n1 is provided to improve the reliability of the memory cell over time, by protecting it against certain types of deterioration appearing with its ageing, as detailed below.

(17) The alternatives C31, C32 of the memory cell C3, represented on FIGS. 19A, 19B, comprise, in the dielectric layer D1, a thicker region D1, situated near the surface of the substrate PW. In the embodiment represented, the thickness of the region D1 increases as it approaches the surface of the substrate and, seen in cross-section, it has a profile substantially in the shape of a V, the dielectric layer D1 having a corresponding profile substantially in the shape of a Y. As a numerical example, the dielectric layer D1 has a thickness of the order of 5 to 10 nm (nanometers), and the region D1 has a thickness of the order of 15 to 20 nm without its portion close to the surface of the substrate.

(18) Furthermore, the lower face of the floating gate FG of the alternative C31 of the memory cell C3 comprises a protuberance p15 which extends beneath the surface of the substrate PW, in the region D1 of the layer D1, and has a face opposite a portion of the selection gate SG.

(19) FIG. 4 shows a preliminary stage of producing the memory cell C3. A deep doped layer NL has been implanted in a semiconductor substrate or wafer WF. This layer is for example the N-type isolation layer of a P-type well forming the substrate PW in which the memory cell is produced. This layer will serve as source line SCL for all the memory cells implanted in the substrate PW, more precisely a collective source plane, capable of collecting the programming currents for programming several memory cells. STI-type (Shallow Trench Isolation) shallow isolating trenches, not visible in FIG. 4 because they are situated in a section plane parallel to that of the figure, may have been made on the surface of the substrate, in the framework of collective manufacturing of several rows of memory cells. A sacrificial oxide layer SOX has then been deposited on the surface of the substrate PW.

(20) During a step shown in FIG. 5, a hard mask HM1 is formed on the oxide layer SOX, by depositing or growing one or more solid layers, for example made of silicon oxide or silicon nitride or a combination of these materials. A light-sensitive resin mask PH is then deposited on the mask HM1, and then developed so as to form an opening 1 in the mask PH.

(21) During a step the result of which is shown in FIG. 6, the mask HM1 has been etched through the resin mask PH so as to form a corresponding opening 1 in the mask HM1, and the mask PH has then been removed.

(22) During a step shown in FIG. 7, the trench 10 is formed in the substrate PW by etching the latter through the opening 1 of the mask HM1. The etching method used is preferably a non-selective and anisotropic dry etching method, such as a plasma etching method. The depth of the trench is here less than the depth of implantation of the doped layer NL. As a numerical example, the trench has a depth of 450 nm and the layer NL is implanted at 750 nm from the surface of the substrate.

(23) During a step shown in FIG. 8A, a deep doped pocket forming the region n0 is implanted in the substrate through the trench 10, in the vicinity of the bottom of the trench 10. The region n0 is formed by vertical ion implantation, and remains localized in the region of the substrate situated in the vicinity of the bottom of the trench 10. The region n0 extends to the doped layer NL and will thus be used as source region for the memory cell being formed, whereas the doped layer NL will be used as source line SCL in the continuation of the source region n0. In one alternative embodiment, the region n0 is not implanted and the trench 10 is etched over a greater depth so as to reach the layer NL, which will be used as source region and source line.

(24) During a step shown in FIG. 8B, a shallow doped pocket forming the floating drain-source region n1 is implanted in a region of the substrate PW extending beneath the hard mask HM1 and the oxide layer SOX. The implantation is made by a tilt angle ion implantation, through the vertical wall of the trench 10 and by using the hard mask as a doping mask, according to a tilt implantation axis shown by an arrow A1. The tilt angle of the implantation axis A1 relative to a vertical axis VA perpendicular to the substrate is chosen so that the region n1 extends beneath the surface of the substrate PW and so that it does not extend too much into the depth of the substrate, to later enable the appearance, between the regions n0 and n1 and opposite the trench 10, of a vertical conductive channel. Present-day technologies offer an implantation angle which can reach 45 relative to the vertical axis VA. This implantation step is here followed by the implantation of the corresponding floating region n1 of the twin cell C3, through the other vertical wall of the trench 10, according to an implantation axis A1 having a tilt angle opposite the angle of the implantation axis A1, relative to the vertical axis VA.

(25) During a step shown in FIG. 9, the dielectric layer D1 is then formed on the walls of the trench 10, for example by oxide growth.

(26) During a step shown in FIG. 10, a conductive layer P1, for example made of polysilicon, is deposited on the entire substrate, as well as inside the trench 10.

(27) During a step shown in FIG. 11, the layer P3 is etched so as not to remain on the surface of the substrate, except inside the trench 10 where it forms the selection gate SG. This step includes the simultaneous etching of the sacrificial oxide layer SOX, or is followed by a step of wet etching the layer SOX.

(28) Next steps, shown on FIGS. 12, 13A, 13B, produce the thicker region D1 in the dielectric layer D1 of the cells C31 (FIG. 19A) and C32 (FIG. 19B).

(29) In the step shown in FIG. 12, a high voltage dielectric layer DHV is deposited on the entire substrate, here by growing a thermal oxide such as silicon dioxide SiO2, for example over a thickness of the order of 10 to 15 nm. This deposit can be carried out in one or more steps and the oxide formed partly comes from the oxidation of the material forming the substrate PW, here silicon. This oxidation generates the region D1 in the vicinity of the surface of the substrate, firstly by oxidation of the material of the substrate which is opposite the trench 10, here silicon, and secondly by oxidation of the material forming the vertical gate SG, here polysilicon. An equal oxidation of the silicon and of the polysilicon on either side of the dielectric layer D1 leads to the substantially V shape of the region D1. The duration of this oxidation step, the thickness of the layer DHV, and the conditions of its implementation enable the depth and the width of the region D1 to be controlled.

(30) It will be noted that this step is optional in relation to the method of manufacturing the memory cell, but can be used in the framework of the simultaneous production of high voltage transistors present in other parts of the circuit into which the memory cell is integrated.

(31) In the step shown on any one of FIGS. 13A, 13B, the dielectric layer DHV is removed by etching, for example by means of a so-called BOE technique (Buffered Oxide Etch) using hydrofluoric acid (HF). This entails a partial removal of the dielectric from the region D1, leading to the appearance of a hollow 15 (FIG. 13A) or a hollow 16 (FIG. 13B) the depth of which depends on the conditions in which this etching step is implemented, and can be controlled by those skilled in the art. Thus, in the examples of embodiments represented, the region D1 of the memory cell C31 being formed (FIG. 13A) has a hollow 15 which extends opposite the embedded gate SG, beneath the surface of the substrate. The hollow is here in the form of a tip, due to the V shape of the region D1. However, the region D1 of the memory cell C32 being formed (FIG. 13B) only has a superficial hollow 16.

(32) In the step shown on any one of FIGS. 14A, 14B, 14C, a tunnel dielectric layer D2, for example silicon oxide, is deposited on the entire substrate, for example over a thickness of the order of 7 to 10 nm.

(33) FIG. 14A shows the profile of the region D1 after depositing the dielectric layer D2 on the memory cell being formed C31. The hollow 15 is only partially filled by the dielectric D2 and still extends opposite the embedded vertical gate SG, beneath the surface of the substrate. FIG. 14B shows the profile of the region D1 after depositing the dielectric layer D2 on the memory cell being formed C32. The superficial hollow 16 is almost completely filled by the dielectric D2. FIG. 14C shows the profile of the layer D2 after it has been deposited on the memory cell being formed C33. The substrate of the memory cell C33, as shown in FIG. 11, has not undergone the step of depositing high voltage dielectric and thus does not comprise the region D1.

(34) In the step shown on any one of FIGS. 15A, 15B, 15C, a conductive layer P2, for example made of polysilicon, is deposited on the entire substrate.

(35) FIG. 15A shows the profile of the layer P2 after being deposited on the memory cell being formed C31. The lower face of the layer P2 has the above-mentioned protuberance p15 which extends in the hollow 15 beneath the surface of the substrate, and has a face opposite a portion of the selection gate SG. FIG. 15B shows the profile of the layer P2 after being deposited on the memory cell being formed C32. The lower face of the layer P2 has a protuberance p16 the extent of which is slight and which extends in the superficial hollow 16. FIG. 15C shows the profile of the layer P2 after being deposited on the memory cell being formed C33. In this embodiment, the lower face of the layer P2 has no irregularity. Below, the protuberance p16 will be considered insignificant and the memory cell C32 will be considered equivalent to the memory cell C33.

(36) FIGS. 16, 17 and 18 show next steps of manufacturing the cell C3. The memory cell represented is the cell C31 but these steps are also applied to the memory cells C32, C33, that are not represented for the sake of simplicity.

(37) In the step shown in FIG. 16, the layer D2 is covered with a dielectric layer D3, for example a so-called interpoly ONO-type oxide (oxide-nitride-oxide). The layer D3 is then covered with a conductive layer P3, here made of polysilicon, and the layer P3 is then covered with a hard mask HM2.

(38) In the step shown in FIG. 17, the hard mask HM2 is etched by photolithography so as to keep only two portions of mask HM2-1, HM2-2 corresponding to the gate stack FG/CG to be produced.

(39) In the step shown in FIG. 18, the layers D2, P2, D3 and P3 are etched by anisotropic dry etching. The regions protected by the portions of mask HM2-1, HM2-2 are not etched and form the gate stack FG/CG comprising the tunnel dielectric layer D2, the floating gate FG, the dielectric layer D3, and the control gate CG.

(40) The positioning of the portions of mask HM2-1, HM2-2 determines the position of the gate stack FG/CG in relation to the vertical gate SG. This positioning is determined when designing the photolithography layout of the memory cell. For this purpose, the designer can define a theoretical overlap distance Dovt corresponding to the targeted overlap distance Dov taking into account a tolerance T of the manufacturing method. The overlap distance obtained Dov is equal to the theoretical overlap distance Dovt plus or minus this tolerance, and thus ranges in the open interval]DovtT; Dovt+T[(the tolerance T considered here being a limit error that is deemed not reached by the manufacturing method).

(41) According to one embodiment, the theoretical overlap distance is Dovt=T, to obtain an overlap distance Dov within the interval]0; 2T[. In other words, the memory cells produced have, between the proximal edge of the gate stack FG/CG and the corresponding proximal edge of the vertical gate SG, an overlap distance Dov ranging from a value close to zero to a value close to 2T, the value close to zero corresponding to an almost perfect alignment of the gate stack with the vertical gate SG.

(42) As an example, with a manufacturing method enabling a floating gate FG to be produced with a length of the order of 120 to 150 nm, a typical tolerance value T is of the order of 20 nm, and the width of the vertical gate SG is of the order of 150 to 300 nm. The overlap distance Dov is then within the interval]0; 40 nm[.

(43) FIGS. 19A, 19B, 19C described above represent the three alternatives C31, C32, C33 of the memory cell C3 obtained after steps of finalizing the manufacturing method, in particular: removing the portions of mask HM2-1, HM2-2, depositing a lateral dielectric layer D4 on the vertical walls of the gate stack FG/DG, and forming in the substrate PW a doped pocket forming the drain region n2, by vertical implantation self-aligned with an external edge of the gate stack FG/CG, this step enabling the drain region n2 of the twin memory cell C3 (C31, C32, C33) to be simultaneously formed.

(44) These steps can be followed by complementary steps that produce a fully integrated circuit. For example, as shown in FIG. 20, the drain regions n2 of the twin cells C3, C3 are then linked to a bit line BL produced in a first level of metal or metal 1, through a conductive via V1 passing through a dielectric layer D5 covering the memory cells. Similarly, the selection gate SG can be linked to a selection line SL formed in a second level of metal or metal 2 through a set of conductive vias V2 passing through the dielectric layer D5 and through a dielectric layer D6 covering the level of metal metal 1.

(45) As shown in FIG. 21, the doped layer NL as source line SCL, here a source plane, can be linked by conductive vias V3 to a set of surface contacts enabling a source line potential to be applied to the layer NL.

(46) FIG. 22 shows a method of programming the memory cell C3 by hot-electron injection. The memory cell represented is the cell C33 but the method is applicable to the other alternatives C31, C32 of the memory cell. The drain region n2 receives a positive drain voltage VD1, for example 4V. The control gate CG receives a positive programming voltage VG11, for example 10V. The selection gate SG receives a positive selection voltage VS1, for example between 1 and 3V. The doped layer NL receives a zero-value source voltage VSC1 (ground of the circuit). The twin memory cell C3, that is connected to the same bit line and thus also receives the voltage VD1, receives on its control gate CG a negative or zero program-inhibit voltage VG12, for example between 2V and 0V.

(47) The bias of the gates CG, SG generates in the memory cell C3 a horizontal channel region CH1 extending beneath the floating gate FG, a vertical channel region CH2 extending opposite the selection gate SG. A current circulates from the drain region n2 to the source region n0 of the memory cell. A flow of electrons EF circulates in the opposite direction to this current, schematically shown by a thick line on the figure. The flow of electrons EF passes through the vertical channel region CH2, then the channel region CH1 to join the drain region n2.

(48) The floating drain-source region n1 is situated at the intersection of the vertical channel region CH2 and of the horizontal channel region CH1. Studies show that the flow of electrons EF bypasses the region n1, as very schematically shown in FIG. 22, because the region n1, due to its N-type doping, weakens the kinetic energy of the electrons and forms a so-called cold region where few electrons circulate.

(49) The channel region CH1 further has an injection zone where hot electrons present in the flow of electrons EF are injected into the floating gate FG, under the effect of a transverse electric field created by the voltage VG11. This region is substantially situated at the intersection of the vertical channel CH2 and of the horizontal channel CH1, near the floating gate FG, and is situated here on the edge of the region n1 since the hot electrons do not cross it.

(50) The doped region n1 reduces the appearance of defects in the memory cell as it ages, or over its cycling (i.e., as it undergoes erasing and programming cycles), and further reduces the effect of such defects on the electrical properties of the memory cell.

(51) Such defects can comprise fixed charges, trapped charges, trapped charges of interface states and other types of electric charges: fixed charges are electric charges appearing at the interface between the substrate PW and the dielectric D1 in response to the electric constraints undergone by the memory cell or induced by oxidation phenomenon, and which may be caused by breaking atomic bonds such as SiH silicon-hydrogen bonds, trapped charges are electric charges stored in the volume of the dielectric layer D1. The appearance of these trapped electric charges can be caused by the breaking of atomic bonds, for example SiO bonds in the case of a dielectric D1 made of silicon oxide SiO2, or by oxygen vacancies in the oxide, trapped charges of the interface states are electric charges trapped by the defects resulting from the discontinuity of the crystalline network, for example at the SiSiO2 interface, electric charges can also be generated by metal impurities or by defects caused by radiation, or even by atomic bond breaking mechanisms. Unlike fixed charges or trapped charges, these last two types of charges can change with the electric potential of the substrate and can greatly influence the electrical characteristics of the memory cell.

(52) In the absence of the region n1, these different defects gradually change the electrical properties of the memory cell, for example its programming or erasing performance, its threshold voltage, its channel length, etc. Thus, for example, the charges trapped in the volume of the oxide can affect the conduction modes in the oxide and reduce the erasing performances of the cell. They are also likely to cause an increase in the threshold voltages in the floating-gate transistor and selection transistor regions, which then causes a decrease in the saturation current and thus a decrease in the injection performance (i.e., slower programming or programming using higher voltages). The trapped charges of the interface states can also act on the conduction properties of the two transistor regions, in particular their slope under the threshold voltage, which also results in a decrease in the injection performance and a deterioration of the read characteristics of the memory cell.

(53) Thus, the region n1 enables a zone of the memory cell that is favorable to the generation of interface states and trapped charges, to be masked, i.e., electrically neutralized. Indeed, this zone is covered with the lateral dielectric layer D4 (FIGS. 19A to 19C) which extends on the edges of the gate stack FG/CG and is likely to receive a nitride layer which can worsen the trapping of charges. Furthermore, in the case of the embodiment shown in FIG. 19B, the region D1 in the shape of a V extending opposite the trench 10, undergoes during the manufacturing of the memory cell a series of oxidation and wet etching operations which can alter its reliability.

(54) As this zone is rendered electrically inactive by the region n1, it no longer takes part in the formation of the channel CH1 or CH2 when the memory cell is put to the conductive state (programming or reading of the memory cell). As schematically shown in FIG. 22, the flow of electrons EF passing through the memory cell is deviated and moved away from this zone. Its possible deterioration thus no longer affects the electrical properties of the memory cell.

(55) The region n1 also locally reduces the intensity of the electric fields during the programming and erasing phases, as the latter can speed up the formation of the aforementioned defects.

(56) On the other hand, providing the region n1 can result in decreasing the injection performance of the memory cell compared to a memory cell which has no region n1. However, this is clearly offset by the advantages offered in terms of stability of its electrical properties over time.

(57) FIG. 23 shows a channel erase method of erasing the memory cell C3. The memory cell represented is the cell C33 but the method is applicable to the alternatives C31, C32 of the memory cell. The region n2 receives a zero-value drain voltage VD2. The control gate CG receives a negative erase voltage VG21, for example 10V. The selection gate SG receives a positive selection voltage VS2, for example 5V. The doped layer NL receives a positive source voltage VSC2, for example 5V. The substrate PW is thus taken to a voltage VB equal to the voltages VS2 and VSC2, for example 5V. Electrons are pulled from the floating gate FG through the substrate and are collected by the source line NL/SCL. The twin memory cell C3 receives on its control gate CG a positive erase-inhibit voltage VG22, for example 2.5V.

(58) FIG. 24 shows an erasing method according to the present disclosure for erasing the memory cell C31, performed through the vertical gate SG. This method is applicable only to the cell C31 and uses the protuberance p15 of the floating gate FG. The region n2 receives a zero-value drain voltage VD3. The control gate CG receives a negative erase voltage VG31, for example 10V. The selection gate SG receives a positive selection voltage VS3, for example 5V. The doped layer NL here receives a zero-value source voltage VSC3. The voltage VB of the substrate PW is thus zero. The twin memory cell C33 receives on its control gate CG an erase-inhibit voltage VG32 that is not necessarily positive and can be zero due to the fact that the voltage VB is itself zero.

(59) Under the effect of the voltage VS3, electrons are pulled from the floating gate FG by the vertical gate SG, and are collected by the word line to which the latter is connected. This process is shown in greater detail in FIG. 25. Between the protuberance p15 and the vertical gate SG extends a dielectric material that is part of the region D1 and which partly comprises dielectric material of the dielectric layer D2, and a composite dielectric material D12 that comprises the original dielectric material of the layer D1 combined with remainders of the high voltage dielectric material DHV. The distance between the protuberance p15 and the gate SG, of the order of a few tens of nanometers, enables the tunnel effect to appear between these two elements.

(60) This gate SG erase method has various advantages compared to the channel erase method. In particular, the erasing is not performed through the same dielectric region as the programming, which decreases the electrical stress of the dielectric material and its ageing. Furthermore, the twin memory cell does not undergo any erase stress (slow spurious erasure) due to the fact that the substrate voltage remains zero, unlike the channel erase method.

(61) It will be noted that the section plane in FIG. 20, and generally speaking the section plane in FIGS. 4 to 19, 22 to 25, is perpendicular to the section plane in FIG. 2. In FIG. 20, the bit line BL is parallel to the section plane whereas the bit line (21) in FIG. 2 is perpendicular to the section plane. Similarly, the selection line SL is perpendicular to the section plane in FIG. 20 and the selection line, formed by the gate material (26), is parallel to the section plane in FIG. 2. Finally, the multiple source bit lines (20) made used by the memory cell structure in FIG. 2 are replaced, for a memory cell structure according to the present disclosure, with the doped layer NL forming a source line SCL and more precisely a source plane for all the memory cells implanted in the same well PW, and capable of collecting the programming currents for programming several memory cells. The memory cell structure according to the present disclosure thus leads to a simpler hot-electron injection programmable memory array architecture than the architecture used by the memory cell structure in FIG. 2.

(62) FIG. 26 shows such a memory array architecture. Only two pairs of twin memory cells C3.sub.i,j, C3.sub.i+1,j are represented, respectively C3.sub.i,j+1, C3.sub.i+1,j+1. The memory cells C3.sub.i,j, C3.sub.i,j+1 belong to a word line WL.sub.i and the memory cells C3.sub.i+1,j, C3.sub.i+1,j+1 belong to a twin word line WL.sub.i+1. The selection gates SG of the selection transistor ST sections of the memory cells are linked to the same selection line SL.sub.i,i+1 and the sources S of the four memory cells are linked to the same source line SCL (formed by the embedded layer NL). The control gates CG of the floating-gate transistor FGT sections of the memory cells C3.sub.i,j and C3.sub.i+1,j are connected to a control gate line CGL.sub.i, and the control gates CG of the memory cells C3.sub.i+1,j and C3.sub.i+1,j+1 are connected to a control gate line CGL.sub.i+1. The drains of the twin memory cells C3.sub.i,j, C3.sub.i+1,j are linked to a bit line BL.sub.j and the drains of the twin memory cells C3.sub.i,j+1, C3.sub.i+1,j+1 are linked to a bit line BL.sub.j+1.

(63) The memory array thus only comprises one bit line per vertical row of memory cells. Each word line WL.sub.i, WL.sub.i+1 only comprises one control gate line CGL.sub.i, CGL.sub.i+1 and one selection line SL.sub.i,i+1 common to the twin word line. The bit line BL.sub.j receives a drain voltage VD.sub.j and the bit line BL.sub.j+1 receives a drain voltage VD.sub.j+1. The control gate line CGL.sub.i receives a gate voltage VG.sub.i and the control gate line CGL.sub.i+1 receives a gate voltage VG.sub.i+1. The selection line SL.sub.i,i+1 receives a selection voltage VS.sub.i,i+1. The common source line SCL, here a source plane, receives the source voltage VSC.

(64) The memory array thus comprises a small number of interconnection lines and its structure is similar to that of a memory array of the type shown in FIG. 1, while benefiting firstly from the advantage offered by a memory cell having a vertical selection transistor section, in terms of footprint, and secondly from the advantage offered by having distinct control and selection gates for the optimization of the programming process.

(65) It will be understood by those skilled in the art that the method according to the present disclosure is susceptible of various other alternative embodiments and applications. In particular, although the description above was of the formation of two twin memory cells, one embodiment of the method according to the present disclosure may be intended to produce unit memory cells, i.e., without any twin memory cell sharing the same vertical selection gate SG.

(66) Conversely, some embodiments may be intended for the collective and simultaneous production of one or more rows of twin memory cells, for example in the framework of the production of an electrically programmable and erasable memory circuit MEM1 of the type shown in FIG. 27.

(67) The circuit MEM1 is fabricated on a semiconductor wafer and forms an integrated circuit IC. It comprises twin word lines WL.sub.i, WL.sub.i+i of the type shown in FIG. 26, fabricated on the substrate PW, two twin word lines WL.sub.i, WL.sub.i+1 comprising twin memory cells sharing the same selection line SL.sub.i,i+1. The selection lines SL and the control gate lines CGL are linked to a word line decoder WLDEC which applies memory cell erase, program and read voltages to them. The bit lines BL connected to the drain regions n2 of the memory cells are linked to a set of programming latches BLT and to a set of sense amplifiers SA through a column decoder CDEC. These elements are linked to a control circuit CCT which ensures the sequencing of programming and erasing operations conforming to one of the methods described above.

(68) It will be understood by those skilled in the art that a memory cell according to the present disclosure is susceptible of being produced in other fields of technology, the materials mentioned in the description above, in particular the silicon, silicon dioxide and polysilicon, being merely examples.

(69) Similarly, the method described above, of forming the region D1 of the dielectric layer D1 and of forming the protuberance p15, is only one example of an embodiment. Other techniques can enable a floating gate FG to be produced that comprises a protuberance enabling the memory cell to be erased through the selection gate. The method described merely has the advantage of not using any additional manufacturing step to produce the protuberance, when it is useful to deposit a high voltage dielectric material on the substrate for the simultaneous production of high voltage transistors.

(70) The various embodiments described above can be combined to provide further embodiments. These and other changes can be made to the embodiments in light of the above-detailed description. In general, in the following claims, the terms used should not be construed to limit the claims to the specific embodiments disclosed in the specification and the claims, but should be construed to include all possible embodiments along with the full scope of equivalents to which such claims are entitled. Accordingly, the claims are not limited by the disclosure.