|Title of Invention||
A SYSTEM FOR HIGH YIELD AND SPEED ENHANCEMENT OF SEMICODUCTOR INTEGRATED CIRCUITS
|Abstract||A novel technique for the enhancement of memory yield and speed of semiconductor integrated circuits using post fabrication transistor mismatch compensation circuitry is proposed. The system is novel because it recognizes that no matter what, the transistor mismatch is statistical in nature and hence it is prudent to exploit the nature of the distribution to get fast and slow circuits rather than make all circuits slow to meet 60" design index. The system comprises of sense amplifier, multiplexer, delay elements, and provision for hardwiring fast and slow circuits during packaging. The sense amplifier firing path is split into slow and fast path and the multiplexer can select one of these. The memory circuits are tested after fabrication to assess whether they could be partitioned as slow or fast circuits and accordingly an appropriate path is selected by the multiplexer. This path is then hardwired during packaging by connecting the select input of multiplexer to VDD or GND.|
|Full Text||FIELD OF TECHNOLOGY
This invention in general relates to manufacturing of semiconductor. Further this invention is related to manufacturing of VLSI circuits. Further this invention pertains to Semiconductor memory technology such as DRAM, SRAM etc. More particularly this invention encompasses a Yield and Speed Enhancement of Semiconductor Integrated Circuits using Post-Fabrication Transistor Mismatch Compensation Circuitry in Semiconductor memory tecimology.
DESCRIPTION OF PRIOR ART:
In semiconductor integrated circuit (IC) manufacturing, there is a typical trade-off between speed and yield. This is even more so when the IC includes analog blocks. For the semiconductor memories such as SRAM and DRAM, ties READ access time determines the speed of the memory. The sense amplifier is the most critical analog block in the READ access path. The response speed of the sense amplifier determines the over all speed of the memory. The typical structure of SRAM memory cell with associated bit line capacitance and sense amplifier is shown in Fig.l. When the cell is accessed, depending on the data stored in the cell, either BL or BL" starts discharging from the pre charged value of Vdd. Since BL and BL" lines have very large capacitance, the sense amplifier is used to amplify the small differential signals on these lines and get the rail to rail swing. The high gain sense amplifier needs to be fired at a precise time when the correct differential voltage is developed at its inputs. Otherwise incorrect data could be latched at the output of the memory.
There are two important factors that impact the sense amplifier firing delay. The first one is related to the tracking of the word line and bit line delays across the process comers and the supply voltage fluctuations. The problem becomes worse with the scaling of CMOS technology into the deep sub-micron regime and the associated lower supply voltages. This problem is typically solved by generating the sense control signal by techniques such as replica memory cell.
The second problem relates to the transistor mismatch in the sense amplifier itself Since the sense amplifier is essentially a differential amplifier, any mismatch in the threshold voltage and the gain factor of the transistors in the two arms of the amplifier shows up as an input offset voltage. In presence of such input offset voltage, the sense amplifier firing signal will have to be further delayed so that the differential voltage developed at the BL and BL" lines can compensate the offset voltage. Since this problem arises AT THE LOCATION of the signal amplification, it is not possible to predict the offset a-priori, or even track the offset as in the case of bit line delay tracking. So the sense amplifier design is typically done for the worst case offset condition, by delaying the firing signal in order to compensate the offset voltage. This approach will slow down the sense amplifier significantiy depending on the magnitude of the mismatch.
As the CMOS technology is scaled to the deep sub micron regime, the random dopant fluctuation effects will increase, thereby making the transistor mismatch worse. If a fab needs a design index of 6a for yield consideration, the sense amplifier firing will be delayed to cancel the 60 transistor mismatch. Assuming a normal distribution of transistor mismatch across the wafer lots, the Fig. 2 gives the percentage of sense amplifiers within certain a value. It should be noted that a la value corresponds to 84% of the devices. In other words, the design index of 6a impUes that for the sake of 16% more devices, 84% of the devices are penalized with respect to the speed. Had it not been for the delayed sense amplifier firing, these 84% of the devices would be in a different higher speed bin, which in turn translates to higher revenue for the fab! At the same time delaying the sense amplifier firing by la value to get high speed IC"s means 16% lower yield. When a fab is shipping millions of IC"s, 16% lower yield implies a big hit on revenue. Hence in order to get the best of speed and yield, one would require a programmable sense amplifier firing design technique.
Some techniques have been proposed to do offset cancellation of the sense amplifier. However, these suffer firom the drawback of slowing down the intrinsic speed of the sense amplifier due to the loading of the output nodes by extra circuitry required for mismatch compensation. Furthermore the mismatch compensation is not to the time of 100%.
1. Typical 6a design index for the fab results in slowing down majority of the chips, which are inherently fast due to lower transistor mismatch. In the Deep Dub-Micron technologies, the speed difference will be more than a factor 2 for the 6a design index versus la design index, i.e. the memory chip which could potentially run at 2GHz, will have to be marketed with a 1 GHz label
2. A manufacturer who likes to be the first to introduce the fastest chips into the market and thereby capture the market can potentially run the fab with la design index. However, the associated loss of yield is enormous which again impacts the bottom line of the fab.
3. Some of the techniques proposed such as cancellation of sense amplifier offset suffer fi-om the disadvantage of slowing down the inherent response speed of the sense amplifier.
OBJECTS OF THE INVENTION:
It is the primary object of the invention to design a novel circuitry to fire the sense-amplifier in a memory circuit wherein, the firing delay introduced is a function of transistor mismatch in the sense ampHfier resulting after the fabrication.
It is another object of die invention to invent a circuitry to extend to have 2 or more speed bins by introducing different delay paths controlled by multiplexer.
It is also another object of this invention that to apply this circuitry to any VLSI circuit wherein the transistor mismatch happens to be a speed and yield critical path.
It is another object of the invention to split the firing delay path into two such as fast path and slow path, which would correspond to lower transistor mismatch and higher transistor mismatch.
It is another object of the invention to adjust the timing for the two paths by a pass gates, buffers, passive interconnect resistor or any other delay element.
It is anotiier object of the invention to use a 2 to 1 multiplexer to select either ^r or slow path using the control signal SPEED to select all the signals of such multiplexers in a single chip to connect together and to bring out the signal to correspond with the SPEED pad.
It is another object of the invention to sort the functional chips as slow chips aadfast chips by testing the functionality with the SPEED pad connected to logic "1" and logic "0".
Now this invention will be described in detail. The description of the invention will refer to the accompanying draAvings of the complete specification. The descriptions will extensively deal with the nature of the invention and the manner in which it is to be performed.
This invention thus provides a system for high yield and speed enhancement of semiconductor integrated circuits such as SRAM and DRAM using post fabrication transistor mismatch compensation circuitry comprising;
(i) a novel sense amplifier for providing hig^ memory yield and speed enhancement using post fabrication, transistor mismatch compensation circuit wherein sense amplifier firing path being split into two path, one path corresponding to fast chips and the other to slow chips;
(ii) a multiplexer which selects one of the two parts of the fire sense amplifier such that more than 80% of the memory integrated circuits (fast path) and the rest of the memory integrated circuits (slow path) are still fimctional without compromising the yield;
(iii) delay elements for effecting delay in the fast and slow path; and
(iv) hardwiring arrangement during packaging.
The statements of the drawings, which accompany this complete specification, are as follows: Fig. 1: Shows the Circuit Diagram of Conventional sense amplifier firing for SRAM
circuit Fig. 2: Shows the graphical representation in respect of the Percentage of devices
within certain mismatch value for Gaussian distribution
Fig. 3(a): Split sense amplifier firing path with multiplexer according to the invaition
Fig. 3(b): Shows the Circuitry indicating Possible delay element and multiplexer
implementation Fig, 4: Shows the details of Hardwiring for fast and slow chips during packaging
Fig No 1 shows the conventional memory design with respect to sense amplifier firing signal (prior art). A Latch type sense amplifier is shown in Fig No 1 here for illustration, the proposed invention coxild be applied to any sense-amplifier configuration. The three pairs transistors MN1-MN2, Mpassl-Mpass2 and MP1-MP2 constitute the sense amplifier. Initially when a memory cell is selected, (WL goes fi-om 0 to 1) the sense amplifier firing signal is low and hence Mpassl and MPass2 are turned on and MNAMP is ofT, thus the sense amplifier is not in regenerating mode. When the memory cell data creates sufficient differential input voltage through Mpassl and Mpass2, the sense amplifier firing signal is enabled (goes fi-om 0 to 1). Now, Mpassl and Mpass2 are turned off, MNAMP is turned on. The 4 transistors MN1-MN2 and MP1-MP2, now act as regenerating latch, thus producing the correct data at the output. If MNl is not matched to MN2, MPl to MP2, and Mpassl to Mpass2, the two arms of the sense amplifier are not balanced which results in a tendency of sense amplifier to lean towards on of the arms, thus creating an input referred offset voltage. Under this circumstance, the sense amplifier can be enabled only when memory data compensates the input referred offset voltage. The input referred offset voltage, if measured on a large number of sense amplifiers fabricated, has a distribution which can be approximated as Gaussian as shown in Fig 2. The process variations result in device mismatch in these matched pairs which manifests itself as the offset of the sense ampUfier. The conventional design (prior art) attempts to accommodate the worst case offset, resulting in pessimistic design by building in slackness to achieve 6a design index. The design slackness is reflected in the delayed sense amplifier firing signal. The 6a design index achieves the high yield (Fig. 2) but at the expense of speed.
The proposed invention is illustrated in Fig 3 (a) and (b). In this proposed invention the sense ampUfier firing path is split into two labeled as la path and 6a path. One of them is corresponding to "fast chips" and the other one to "slow chips". The delays in the fast and slow paths are appropriately adjusted depending on the required design index. For the example shown in the Fig. 3 (a), the fast path corresponds to la and the slow path
corresponds to 6a design index (transistor mismatch). These niraibers could be different depending on the requirement of a particular implementation. The delays in these two paths can be tuned by using delay elements such as pass gates, buffers, passive interconnect resistors etc as shown in Fig. 3(b). The delay in pass gate and buffer can be adjusted by varying the length and width of the transistors whereas the delay in the intercoimect can be adjusted by varying length and width of metal line. The implementation of 2-1 multiplexer is shown in Fig. 3(b). The 2-1 multiplexer selects one of the two paths (In-1 and In-2) to fire the sense amplifier (Out) as shown in Fig. 3(a). The status of select input (GND or VDD) of the multiplexer determines whether In-1 or In-2 is transmitted to the output. The select inputs of all such multiplexers in a single chip are connected together and brought out as the "SPEED" pad as shown in Fig. 4. When "SPEED"=0, liiefast path is used to control sense amplifier and when "SPEED"=1, the slow path is used. (The logic could also be reversed, by changing the multiplexer connections appropriately). After the fabrication is completed, the IC"s are tested at wafer level to assess the functionality. If a particular chip works with SPEED=0, then it is marked as fast chip. Otherwise, the chip is re-tested with SPEED=1, and the functional chip is marked as slow chip. Then after dicing, the chips are hardwired into fast or slow bins during the packaging step. For all the^^ chips, the SPEED pad is shorted to GND pad and for the slow chips the SPEED pad is shorted to VDD pad as shown in Fig. 4. The VDD and GND pads are adjacent to SPEED pad. Note that the SPEED pad need not be connected to an external package pin and hence the hardwiring process does not impact the comiectivity of the chip in the system design.
This results in 84% of the chips in high speed bin and 16% of the chips in low speed bin without compromising on memory speed and yield. This in turn translates into a very significant enhancement in the revenue of the fab.
While the invention has been particularly shown and described with reference to the preferred embodimoits thereof, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing firom the spirit and scope of the invention.
1. A system for high yield and speed enhancement of semiconductor integrated circuits
such as SRAM and DRAM using post fabrication transistor mismatch compensation
(i) a novel sense amplifier for providing high memory yield and speed enhancement using post fabrication, transistor mismatch compensation circuit wherein sense campfire firing path being split into two path one path corresponding to fast chips and the other to slow chips;
(ii) a multiplexer which selects one of the two paths to fire the sense amplifier such that more than 80% of the memory integrated circuits work as fast circuits(fast path) and the rest of the memory integrated circuits are still functional (slow path) without compromising the yield;
(iii) delay elements for effecting delay in the fast and slow path; and
(iv) hardwiring arrangement during packaging.
2. A System for high yield and speed enhancement of semiconductor integrated circuits as claimed in Claim 1, wherein a split firing path one corresponding to "fast chips" and the other "slow chips" are adjusted depending on the required design index.
3. A System for high yield and speed enhancement of semiconductor integrated circuits as claimed in Claim 1, wherein the fast path corresponds to No and the slow path corresponds to Mo design index (transistor mismatch) wherein these numbers could be different depending on the requirement of a particular implementation thus wherein M>N.
4. A System for high yield and speed enhancement of semiconductor integrated circuits as claimed in Claim 1, wherein the delay in two paths can be timed by using delay elements such as gates, buffers passive interconnect resistors.
5. A System for high yield and speed enhancement of semiconductor integrated circuits
as claimed in Claim 1, wherein the 2-1 multiplexer selects one either fast or slow path
using the control signal SPEED thus all control inputs of such multiplexers in a single
chip are connected together and brought out which corresponds to the "SPEED" pad.
6. A System for high yield and speed enhancement of semiconductor integrated circuits as claimed in Claim 1, wherein the fast path used when SPEED=0 and the slow path is used when SPEED=1 thus the logic could also be reversed, by changing the multiplexers connections appropriately.
7. A System for high yield and speed enhancement of semiconductor integrated circuits as claimed in Claim 1, wherein after the fabrication is completed, the chips are tested at wafer level to assess the functionality, and marking a particular chip as the fast chip, if the chip is functional with SPEED=0 and slow chip when it does not function with SPEED=0 but functions with SPEED=l, thus dicing and separating into fast or slow bins for hardwiring during the packaging.
8. A System for high yield and speed enhancement of semiconductor integrated circuits
as claimed in Claim 1, wherein for all the fast chips, the SPEED pad is shorted to
GND pad and for all the slow chips the SPEED pad is shorted to VDD pad wherein
the VDD and GND pads adjacent to SPEED pad.
9. A System for high yield and speed enhancement of semi conductor integrated circuits,
substantially as herein before described and illustrated in figures 2, 3(a), 3(b) and 4.
|Indian Patent Application Number||216/MAS/2001|
|PG Journal Number||20/2006|
|Date of Filing||12-Mar-2001|
|Name of Patentee||M/S. INDIAN INSTITUTE OF SCIENCE|
|Applicant Address||BANGALORE 560 012|
|PCT International Classification Number||H03K 5/00|
|PCT International Application Number||N/A|
|PCT International Filing date|