333. The memory cells 333 are coupled to sense lines 335 and access lines 336. In some examples, memory array 310 may be included in memory array 110 and/or memory array 210.

In some embodiments, to implement a Bloom filter, the memory array 310 may act as a multiply-accumulator (MAC). A resistive element 334 of each memory cell 333 may be programmed to have a desired resistance (R). As used herein, the terms resistance and resistance values are used interchangeably. Voltage (V) inputs may be provided along each access line 336. As used herein, the terms voltage and voltage values are used interchangeably. The resulting current through each memory cell 333 due to the application of the voltage inputs is provided to a sense line 335, which may provide a current output. The voltage inputs may be provided by DACs, such as DACs 114 and/or 214 in some examples. The current output may be provided to sense circuitry such as sense circuitry 111 and 211 in some examples.

In the example shown in FIG. 3, a voltage V1 is provided along access line 336A to a memory cell 333A having a conductance G1 and a voltage V2 is provided along access line 336B to a memory cell 333B having a conductance G2. As is understood in the art, the conductance of the memory cells 333 may be based on the resistance of resistive elements 334. The resulting current I1 through memory cell 333A is equal to V1*G1. Current I1 is provided to sense line 335. The resulting current I2 through memory cell 333B is equal to V2*G2 is also provided to sense line 335. The total current I on sense line 335 is the sum of I1 and I2. Thus, the total current is the sum of the multiplication operations at each of the memory cells 333 along the sense line 335 to implement the multiply-accumulate operation:

I
=

I
⁢
1

+

I
⁢
2

=

(

V
⁢
1
×
G
⁢
1

)

+

(

V
⁢
2
×
G
⁢
2

)

Equation
⁢

1

Thus, the total current for a sense line 335 is a sum of all the products of the voltage values along the access lines 336 and the conductance values of the memory cells 333 along the corresponding sense line 335. As is understood in the art, if the states of the memory cells 333 are represented as resistance (R), where G=1/R, the total current I would be the sum of division operations:

I
=

I
⁢
1

+

I
⁢
2

=

(

V
⁢

1
÷
R

⁢
1

)

+

(

V
⁢

2
÷
R

⁢
2

)

Equation
⁢

2

Although only one sense line 335 and two memory cells 333 are shown in FIG. 3, the voltage inputs along access lines 336 may be provided to memory cells along multiple sense lines simultaneously, or near simultaneously. Multiple sense lines may provide outputs simultaneously or near simultaneously. Thus, multiple multiply-accumulate operations may be performed in parallel in the memory array 310.

According to embodiments of the present disclosure, the resistance levels and voltage levels can represent data, such as data that corresponds to a nucleotide of a genetic sequence. For example, each nucleotide type A, C, G, T, and/or U may correspond to a different level (e.g., value) of resistance (e.g., conductance) and/or voltage. A genetic sequence may be stored in one or more memory cells 333 along a sense line 335 of the memory array, and another genetic sequence may be provided as an inputs along access lines 336. The stored genetic sequence may be stored as a set of resistances, and the input genetic sequence may be provided as a set of voltages. This may result in a current output along sense line 335. The resulting current may be used to determine whether the input genetic sequence possibly matches the genetic sequence stored in the memory cells 333 stored along sense line 335.

FIG. 4 illustrates an example of a multiply-accumulate operation for a genetic sequence in accordance with a number of embodiments of the present disclosure. Genetic sequence 440 includes multiple nucleotides 444. In this example, genetic sequence 440 is a DNA sequence that includes nucleotides A, C, G, and T, but in other examples, genetic sequence 440 may be an RNA sequence or other genetic sequence. Each nucleotide type is assigned a different resistance/conductance value and a different voltage value compared to the other nucleotide types. In the simple example shown in FIG. 4, each nucleotide type is assigned a same value for both resistance/conductance and voltage as shown in legend 442. In other examples, each nucleotide type may be assigned different values for resistance and voltage. Further, although values 1, 2, 3, and 4 are used in the present example, other values may be used in other examples. For example, prime numbers (e.g., 1, 3, 5, and 7) may be used.

The genetic sequence 440 is multiplied with itself on a nucleotide-by-nucleotide basis. In the example shown in FIG. 4, the multiplication operations are G×G (3×3), A×A (1×1), T×T (4×4), T×T (4×4), A×A (1×1), C×C (2×2), and A×A (1×1). The voltage assigned to each nucleotide 444 of sequence 440 is multiplied with the conductance assigned to the corresponding nucleotide 444 of sequence 440 to provide a resulting current value 446 for each nucleotide 444. The resulting currents for all of the nucleotides 444 in the sequence 440 are then summed (e.g., accumulated) to provide a total current 448. The total current in the example shown in FIG. 4 is 48 (e.g., uAmp, mAmp, etc.).

The resulting total current 448 indicates an expected total current (e.g., target value) from a multiply-accumulate operation when an input genetic sequence matches a genetic sequence stored along a sense line (e.g., sense line 235, 335) of a memory array (e.g., memory array 110, 210, 310). When the resulting current of a sense line does not match the target value for an input genetic sequence, the genetic sequence stored along the sense line may be definitively determined not to be a match of the input sequence. When the resulting current of the sense line does match the expected resulting current for the input sequence, it indicates a potential match. The input sequence provided to the memory array and the genetic sequence stored along the sense line of the memory array may be compared by another technique to confirm the match. Thus, a memory array performing multiply-accumulation operations may act as a Bloom filter.

FIG. 5 illustrates an example of storing a reference sequence in a memory array in accordance with a number of embodiments of the present disclosure. The memory array 510 may be included in memory array 110, 210, and/or 310 in some embodiments. The memory array 510 may include memory cells 533 along sense lines 535 and access lines 536. Memory cells 533 may be implemented by memory cells 233 and/or 333 in some examples. Sense lines 535 may be implemented by sense lines 235 and/or 335 in some examples. Access lines 536 may be implemented by access lines 236 and/or 336 in some examples. While for illustration, memory array 510 is shown including seven access lines 536, it is understood that memory array 510 may include any number of access lines 536. Furthermore, although only four sense lines 535 are shown in FIG. 5, memory array 510 may include any number of sense lines 535. While each memory cell 533 may include a resistive element (e.g., resistive element 234 and/or 334) or other element capable of performing a multiply-accumulate operation, for ease of illustration, the memory cells 533 are illustrated as blocks with letters indicating the nucleotide stored in the memory cell 533 (e.g., based on a resistance programmed in the memory cell 533).

A reference sequence 550 may be provided to the memory array 510 (e.g., via control circuitry 105 and/or I/O circuitry 107) for storage in the memory array 510. A portion of the reference sequence 550 may be stored along each sense line 535. In some examples, a sliding window 501 having a length of the sense line 535 may be progressed along the reference sequence 550, and each sense line 535 may store the portion of the reference sequence 550 located within the sliding window 501 at a particular position in the progression. In some examples, such as the one shown in FIG. 5, the sliding window 501 may be progressed one nucleotide at a time, and the portion of the reference sequence 550 stored in a sense line 535 may be shifted by one nucleotide compared to the adjacent sense line 535. However, the sliding window 501 may be progressed by more than one nucleotide in other examples.

In some examples, the sliding window 501 may be progressed until the end of the reference sequence 550. If the reference sequence 550 is long enough such that the memory array 510 “runs out” of sense lines 535 to store portions of the reference sequence 550, the remaining portions of the reference sequence 550 may be stored in another memory array. In some examples, the remaining portions may be stored in a memory array of a same bank as memory array 510. In some examples, the remaining portions may be stored in a memory array of a different bank as memory array 510.

While one reference sequence 550 is shown in FIG. 5, in some examples, multiple reference sequences may be stored in memory array 510. In some examples, different reference sequences may be stored in different memory arrays of a bank including memory array 510. In some examples, different banks may store different reference sequences. In some examples, how the reference sequence 550 is divided into portions, stored, and/or shifted in the memory array 510 (or across multiple memory arrays) is performed based on control signals provided by control circuitry, such as control circuitry 105. In some examples, the control signals may be based, at least in part, on instructions/commands provided by a host, such as host 102.

In some embodiments, once the reference sequence 550 is stored in memory array 510 based, at least in part, on resistance values of the memory cells 533, reads may be provided to memory array 510 for comparison to the reference sequence 550.

FIG. 6 illustrates an example of comparing reads to reference sequences stored in the memory array shown in FIG. 5 in accordance with a number of embodiments of the present disclosure. In some examples, read 652 may be provided as voltages along access lines 536 (e.g., via DACs 114, 214). For ease of illustration, rather than voltage levels along the access lines 536, read 652 is illustrated as blocks with letters indicating the nucleotide represented by the voltage level. In some examples, data indicating the voltages to be provided for the read 652 may be stored in a register of the memory device including memory array 510. In some examples, the data indicating the voltages to be provided for the read 652 may be provided in a different memory array and/or memory bank of the memory device. In some examples, the data indicating the voltages to be provided for the read 652 may be provided by control circuitry and/or I/O circuitry of the memory device (e.g., control circuitry 105 and I/O circuitry 107).

In some examples, the read 652 may be the same length as the portions of the reference sequence 550 stored along each sense line 535. In some examples, the length of the portions of the reference sequence 550 stored along each sense line 535 (e.g., the length of the sliding window) may be based, at least in part, on an expected shortest read generated by a sequencing system. In some examples, read 652 may be “padded” by values for voltages that indicate a “don't care” and/or that will not affect the results of the multiply-accumulate operation.

The read 652 may be provided to all of the portions of the reference sequence 550 along each of the sense lines 535 simultaneously or near simultaneously. The voltage corresponding to the nucleotide along each access line 536 may pass through the corresponding memory cells 533 of the access line 536, generating a current. The current may be based on the voltage provided to the access line 536 and the resistance/conductance value programmed to each memory cell 533. The current through each memory cell 533 may be provided to the corresponding sense line 536. The currents along the sense line 536 may be summed, thus a multiply-accumulate operation may be performed along each sense line 536. The sum of the currents of the memory cells 533 may be provided from the sense lines 536 to sense circuitry (e.g., sense circuitry 111, 211). In some examples, the sense circuitry may provide current outputs for analysis. In some examples, the sense circuitry may provide the sensed currents to ADCs (e.g., ADCs 115, 215) that may convert the currents into digital signals representing the currents as an output.

For ease of illustration, the outputs 656 associated with each of the sense lines 536 are illustrated as boxes with numerical values reflecting the result of the multiply-accumulate operations of the read 652 with the portions of the reference sequence 550 based on the values provided in legend 442. As described with reference to FIG. 4, the target value 654 (e.g., expected value) of the read 652 multiplied and summed against itself is 48. Several outputs 656 have values that are not equal to 48. These outputs are true negatives 658. In some embodiments, there may be no mechanism for the MAC of the Bloom filter as disclosed herein to provide a false negative. Accordingly, the portions of the reference sequence 550 stored along sense lines 535 associated with the true negatives 658 may be ignored or discarded as potential matches for the read 652.

While there may not be false negatives, there may be multiple combinations of conductance and voltage values that when multiplied and summed equal 48, which may lead to false positives. True positive 660 output correctly indicates a portion of the reference sequence 550 of a sense line 535 matches the read sequence 652. However, false positive 662 also equals 48, even though the portion of the reference sequence 550 stored along the associated sense line 535 does not match read sequence 652. Accordingly, outputs 656 that match the target value 654 may be further analyzed prior to confirming a potential alignment location of the read 652 in the reference sequence 550. In some examples, the portions of the reference sequence 550 stored along the sense lines 535 having positive results (e.g., outputs 656 match the target value 654) may be compared to the read 652 on a nucleotide-by-nucleotide basis. In some examples, other comparison techniques may be used. In some examples, this additional analysis may be performed on the memory device, such as by pattern matching circuitry, which may include a content addressable memory and/or one or more comparator circuits. In some examples, the portions of the reference sequence 550 and the read 652 may be provided to an external device, such as host 102, which may perform the additional analysis. This additional analysis may filter out false positive 662 and retain true positive 660.

Once confirmed, the true positive 660 indicates that the read sequence 652 may align to the reference sequence 550 at the location of the reference sequence 550 along the corresponding sense line 535. Because it is known which portion of the reference sequence 550 is stored along each sense line 535, it can be determined where in the reference sequence 550 read 652 may be aligned to. In some examples, such as when the memory confirms the true positive 660, the memory may provide an output indicating a location in the reference sequence 550 that the read 652 aligns to (e.g., alignment location). The output may be stored in a register or memory array of the memory device, such as memory device 103. In some examples, the output may be provided to an external device, such as host 102. In other examples, such as when the external device performs the additional analysis, the alignment location may be provided to the memory device or another memory device for storage and/or used in further processing for recombining all of the analyzed reads into a longer sequence (e.g., a sequence of the sample from which the reads were acquired).

While positive results 660, 662 may require additional computation time to confirm, the ability of the Bloom filter implemented by memory array 510 to compare the read 652 to many portions of the reference sequence 550 in parallel and definitively eliminate portions that are not matches, the overall computation time for aligning the read 652 to the reference sequence 550 (or determining the read 652 does not align to the reference sequence 550 when there are no matches) may be less.

While read 652 is shown as being compared against multiple portions of a reference sequence 500 in parallel in a single memory array 510 in FIG. 6, in some examples, read 652 may be provided in parallel to multiple memory arrays and/or banks of memory arrays in parallel. Each memory array and/or bank may include one or more reference sequences. In some examples, reads may be provided to the memory device in series (e.g., one read at a time is compared to the reference sequence(s) in the memory array(s)). However, in some examples, multiple reads may be compared to reference sequences in parallel. For example, different reads may be provided to different memory arrays and/or banks in parallel.

As described with reference to FIG. 6, any output 656 that does not match the target value 654 is discarded or ignored. However, a genomic sample from which reads are obtained does not typically have a sequence that exactly matches a reference sequence, even if the genomic sample is a same organism. For example, mutations may cause changes in the sequence between the sample and the original organism from which the reference sequence was obtained (e.g., alpha, beta, delta, and omicron variants and sub-variants thereof for COVID-19). Additionally, the process of obtaining reads from the sample is not perfect. Some reads may include one or more of a mismatched nucleotide (e.g., a transcription error), a deletion of a nucleotide and/or an insertion of a nucleotide.

In some applications, if only outputs 656 that exactly match the target value 654 are retained/further analyzed, the Bloom filter may not tolerate mutations or read errors. This may lead to an unacceptable number of reads marked as not aligning to the reference sequence. To increase error tolerance, in some embodiments, true negatives 658 with values within a range of the target value 654 may be retained/further analyzed. For example, true negatives 658 having a value within +/−2 of the target value 654 (e.g., between 46 and 50) may also be analyzed as a possible “close match.” Further analysis performed by the memory and/or host may determine whether the close matches are false close matches or “true” close matches due to a mutation and/or read error. The width of the range may be based, at least in part, on a length of the read 652 and/or a desired error tolerance. Increasing the error tolerance in some cases may increase computation time as it may lead to additional portions of the reference sequence 550 requiring additional analysis (e.g., nucleotide-by-nucleotide comparison and/or additional processing to determine error/mutation type).