Description
Semiconductor Device
Technical Field
The present invention relates to a semiconductor device, and in particular to a high-performance CMOS circuit.
Background Art
Source-follower circuits are frequently used to drive impedance loads, especially in applications employing analog or multiple- valued signals. Such a circuit is depicted in Fig. 1. This diagram indicates a source-follower circuit comprised of one NMOS transistor (abbreviated as "NMOS") (Nl) and a load capacitance (CL); when Nl is turned on, the current ID increases VOUT until
V0UT = VIN - VT, (1) where VT is the threshold voltage of the NMOS. VT is given by:
vτ = Vτo + Y (V2ΦF + |V SB I - 2ΦlF )> (2)
where γ is the body effect coefficient, VSB is the differential source-body voltage, and ΦF is defined by:
|Ef - Ei | ΦF = — -■ (3) q
Ef is the Fermi level of the semiconductor, E\ is the intrinsic Fermi level, and q is the charge of an electron.
A problem of this circuit arises if the output voltage is desired to be equal to the input voltage. In the case of a source-follower circuit, if the body of the NMOS is biased at 0 V, then VSB will increase as the output voltage rises. This will cause VT to rise, even if Vχo = 0 V. This phenomenon is called the "body
effect". Therefore, VoUT is always less than VI for a source follower containing an enhancement-mode NMOS.
Another drawback of the source-follower circuit is its slow transient characteristics. As VoUT rises, the transistor's gate-source differential voltage VQS (= VI - VOUT) decreases, causing a reduction in the channel conductivity, which in turn reduces the drain current ID- Therefore, the amount of current available to pull up VoUT continuously decreases. This results in the transient characteristics shown in Fig. 2.
The present invention was created in order to solve the problems stated above; it has as an objective thereof to provide a semiconductor device which makes possible the full restoration of the source-follower output to equal its input voltage, and also decreases the time required for the output to reach its maximum value. In addition, this invention can also be used to convert the analog output of a source-follower circuit into a digital form.
Disclosure of the Invention
The present invention discloses a semiconductor circuit comprising at least an MOS transistor. A multiple-valued data line, which is capable of M discrete voltage levels, or an analog data line rises (or falls) from an initial voltage to its final voltage. The inputs of M-l comparators are connected to said data line, the outputs of said comparators are coupled capacitively to the input gate of a source-follower circuit, and the output of said source-follower circuit is fedback to the data line.
By means of the above semiconductor circuit, the time necessary for the data line to reach its final voltage is drastically reduced. Furthermore, the voltage of the data line can be easily converted to a binary-digital form with this circuit.
Brief Description of the Drawings
Figure 1 is a conceptual diagram showing a typical NMOS source-follower circuit.
Figure 2 is a graph showing time dependence of the output voltage of an NMOS source follower.
Figure 3 is a conceptual diagram showing the circuit of Embodiment 1.
Figure 4 is a graph showing the simulation data of the circuit of Embodiment 1.
In fugyre 4 SAB is sence and boost.
Figure 5 is a conceptual diagram showing the circuit of Embodiment 2.
Figure 6 is a conceptual diagram showing the circuit of Embodiment 3.
Figure 7 is a conceptual diagram showing the circuit of Embodiment 4.
Figure 8 is a conceptual diagram showing the circuit of Embodiment 5.
Figure 9 is a conceptual diagram showing the circuit of Embodiment 6.
Best Mode for Carrying Out the Invention
Herein below, the present invention will be explained in detail based on embodiments; however, the present invention is of course not limited to these embodiments.
(Embodiment 1)
This invention is used in a multiple-valued system containing M voltage levels. The lowest voltage level (Level 0) is determined to be equal to Vss, and the highest voltage level (Level M-l) is determined to be equal to V D, where
VDD and Vss are both power supplies to the system, and VDD i a higher voltage than Vss- In general, the voltage corresponding to Level x is v ve,,. -(VD° : ss) -χ ÷vss (4)
The first embodiment of the present invention is shown in Fig. 3. In the diagram, an M-valued data line, which is the output of an NMOS source- follower circuit (N2), is connected to the inputs of M-l analog comparators. The outputs of these comparators are capacitively coupled to the floating gate of a large NMOS (N3). The drain of N3 is connected to VDD, the source of N3 is connected to the M-valued data line, and the body of N3 is connected to Vss- N3 is designed to be much larger than N2.
In this circuit, the comparators simultaneously compare the voltage of the M-valued data line (VDATA) to the voltages corresponding to Level 0 through Level M-2. If VDATA rises beyond V evel 0 , then the Level-0 comparator output will become VDD- If VDATA continues to increase above VLevel 1, then the Level- 1 comparator output will also become VDD, and so forth. As the comparators turn on one by one, the potential (Φpi ) of the floating gate of N3 increases. Since the comparator outputs (Vco, Vci, Vc2? —, Vc(M-2)) have equal coupling capacitances to the floating gate, and said coupling capacitances are designed to be much larger than the coupling capacitance from the floating gate to the body of N3, ΦFI is given by the following equation:
V cO + Vci + VC2 + • • • + VC(M-2)
Since N3 is also in a source-follower configuration, VDATA will rise until it equals Φpi - VτN3-
The function of this invention is to continuously detect the present level of VDATA and then boost it to the next higher level. It continues the cycle of
detecting and boosting until current has stopped flowing from N2 (i.e., until N2 is cut-off). In this manner, the capacitance load on the data line is primarily pulled up by the large current flowing from N3. Therefore, the data readout from the source follower of N2 is drastically accelerated, and the problem of slow source-follower circuit operation has been resolved.
In the present invention, the body effect factor γ can be reduced by connecting the body of N3 to a voltage source lower than Vss; that is, a negative substrate bias may be applied to the p-type substrate. Furthermore, if the body of N3 were an isolated well connected to the data line, then the second term in Eq. 2 would equal 0. Such isolation can be realized by forming p-wells in an n-type substrate for each N3 transistor, or by using a silicon-on-insulator substrate. The elimination of the body effect in this manner would make the gain of this circuit equal to one. In this case, VDATA would rise until it equals
ΦFI -
In this embodiment, the coupling capacitances of the comparator outputs to the floating gate of N3 were all made equal. However, they can be designed to have different values, and the effective gain of the N3 source follower can be modified if necessary.
This invention may be used in an analog system if the voltages of the analog data line could be classified into M discrete levels.
Furthermore, the outputs of the comparators are in digital form, thus providing a means to simply convert the multiple-valued or analog signal into a digital signal.
Figure 4 displays the HSPICE simulation of the output characteristics of 4-valued NMOS source-follower circuits. The dashed lines show the output characteristics without the present invention, and the solid lines show the output
characteristics with the present invention. It is clearly seen that the present invention drastically accelerates the source-follower readout.
(Embodiment 2)
If the body of transistor N3 in Embodiment 1 were connected to the data line, then the junction capacitance of the well containing N3 would be added as a parasitic capacitance to the data line. Since N3 should be designed to be very large, the parasitic junction capacitance of its well would also be very large.
To offset this disadvantage, the circuit in Fig. 5 is presented in this embodiment. This circuit is similar to that of Embodiment 1 , except that N3 is divided into two transistors connected in parallel: N3A (whose body is biased to Vss) and N3B (whose body is connected to the data line). In this manner, N3A, with no large parasitic junction capacitance, aids in pulling up VDATA and N3B will overcome the body effect of N2 and N3A. The ratio (W/L)N3A:(W/L)N3B must be optimized for speed according to the device parameters.
In Embodiments 1 and 2, the invention is applied to speed up the voltage transition of the data line from an initial voltage to a higher voltage. This is because the data line is initially driven by a source follower consisting of a small NMOS transistor. For this reason, a large NMOS transistor was used to raise the data line. Conversely, if the data line is driven by a small PMOS source follower from a higher initial voltage to a lower voltage, then a large PMOS transistor should be used to speed up the process.
So far, the explanations have been provided for cases where the initial data line transient is caused by a MOS source-follower circuit. However, this invention is not limited to these cases. The data line can be initially driven by a resistor directly connected to a voltage source or by an emitter-follower circuit
of a bipolar transistor, for example. In any case, this invention is applicable to speed up a transient by using sensing and boosting circuitry consisting of MOS source followers.
(Embodiment 3)
In the previous embodiments, N3 can be replaced by a CMOS source- follower circuit, consisting of an NMOS (N4) and a PMOS (P4), as depicted in Fig. 6. Unlike N3, however, the output of the CMOS source-follower circuit is constrained to equal ΦFI ; any current flowing from N2 would find a conducting path to Vss through P4. Therefore, a switch must be added between the CMOS source-follower output and the data line. The switch must be off initially to allow N2 to charge up the data line, and the switch must be on whenever the CMOS source follower is charging up the data line; the switch must be turned off again whenever the comparators are detecting whether or not current is still flowing from N2.
(Embodiment 4)
If the body effect observed in the NMOS source follower of Embodiments 1 and 2 are appreciable, then the data line cannot fully reach the maximum value of VDD- The circuit shown in Fig. 7 overcomes this problem by using a PMOS transistor (P5) as a switch between VDD and the data line; the output of the Level-(M-l) comparator is connected to the input of an inverter, and the output of said inverter is connected to the gate electrode of the PMOS. Therefore, this PMOS turns on when the highest level is sensed, i.e., when the Level-(M-l) comparator turns on. In this manner, the data line is able to rise to the maximum value of VDD when reading out the highest voltage level, thus giving the source follower an effectively unity gain.
Similarly, if a PMOS source follower is used in Embodiments 1 and 2, then the data line cannot fully reach the minimum value of Vss- This problem can be similarly solved by using an NMOS transistor as a switch between the data line and ground; the NMOS turns on when the lowest level is sensed.
Furthermore, if the CMOS source follower of Embodiment 3 is used, then the loss of gain resulting from the body effect is compensated by using a PMOS switch between VDD and the data line, and an NMOS switch between Vss and the data line. The PMOS turns on when the highest voltage level is sensed, and the NMOS turns on when the lowest level is sensed.
(Embodiment 5)
The fifth embodiment of the present invention is shown in Fig. 8. This circuit depicts the Level-x comparator used in Embodiments 1 -4. The comparator has two inputs, V adjust and VDATA, which are capacitively coupled to a floating gate electrode. Said floating gate is the input gate of a first CMOS inverter. This inverter contains a PMOS (P6), and an NMOS (N6). The output of said first inverter is connected to a second CMOS inverter consisting of P7 and N7. The output of said second CMOS inverter serves as the output of the
Level-x comparator, Vex- Let the inversion threshold of both inverters be v (Vpp - V ss ) A v ,.
ViNV j + VSS- (6)
The coupling capacitances of Vadjust and VDATA are Ci and C2, respectively. Then the potential ΦF2 of the floating gate is
if COP, CON « CI , C2.
Here, COP and CON are the coupling capacitances from the floating gate to the substrates of P5 and N5.
Since the purpose of this circuit is to detect when VDATA surpasses VLevel x, it must operate under the following conditions:
If VDATA > VLevel x, then Φ 2 > Vi V; therefore Vex = VDD-
If VDATA < VLevel x, then ΦF2 < VINV; therefore Vex = Vss-
These conditions are achieved by adjusting the coupling capacitance ratio :C2 or the input voltage Vadjust-
(Embodiment 6)
In Embodiment 5, the bias voltage of Vadjust was directly supplied to an input gate in order to control the effective inversion threshold as seen from VDATA- However, the same effect may be achieved by breaking up control gate Ci into multiple gates and connecting said gates to either VDD or Vss, as shown in Fig. 9. In this manner, only two different power supplies are required.
(Embodiment 7)
In Embodiments 1 through 6, the floating gates are completely electrically isolated. However, it is possible that over time, some amount of charge could be injected into the floating gate. In order to reset the floating gate so that there is no net charge, a switch may be added between the floating gate and ground (or a voltage source which sets the initial condition). This switch is off when the circuit is in operation but turns on during a refresh cycle. For refresh operation, all input terminals capacitively coupled to the floating gate need to be grounded or set to appropriate initial potentials in order to guarantee the same initial condition. However, these potentials, as well as the floating gate's refresh potential, may be changed at each refresh cycle to any value.
This invention is especially suitable for multiple-valued memories which use a source-follower gain cell. In such an application, several memory cells are connected to one bitline. Since the bitline capacitance is large, it would take a very long time for one memory cell to raise the bitline voltage to the value stored in the cell. If the present invention is connected to the bitline, it would drastically accelerate the readout time of the memory cells. In this manner, the invention would serve as a sense amplifier for the memory cells If the memory cells are dynamic, then this invention would also serve as refresh circuitry.
Additionally, this invention is not limited to readout of source-follower circuits. It can be applied in any system where a multiple- valued or analog data line is employed.