A proper understanding of logic density is important for designers when they compare the size of programmable logic devices (PLDs) from different vendors. Unfortunately, different PLD vendors publish logic-density data with different terminology and counting schemes, which complicates the device-density selection process. The largest PLD densities in the market today are from the Altera APEX 20KE and Xilinx Virtex-E families. This web page provides designers with a calculation method that effectively compares the densities of the Altera APEX 20KE and Xilinx Virtex-E device families.
Counting Basic Building Blocks
The most direct method of measuring the logic density of PLDs is to count the number of basic building blocks. The basic building block of Altera APEX 20KE devices is the logic element (LE). One LE contains a 4-input look-up table (LUT), a register, and additional carry and cascade logic. The basic building block of Xilinx Virtex-E devices is the logic cell (LC). According to the Xilinx Virtex-E data sheet, "An LC includes a 4-input function generator [LUT], carry logic, and a storage element [register]" (p. 5). Comparing these basic building blocks makes it easy to calculate logic density. Table 1 provides a comparison of the APEX 20KE and Virtex-E architecture terms.
| Table 1. PLD Architecture Definitions | ||
|---|---|---|
| Term | Definition | |
| APEX 20KE | Virtex-E | |
| LE | LC | The basic building blocks of a PLD, containing a 4-input look up table (LUT), a register, and additional logic |
| 4-Input LUT | 4-Input Function Generator | An SRAM-based logic function that can implement any functions with up to 4 inputs |
| Programmable Register | Storage Element | A register that can be configured as various types of flipflops |
| Carry Logic | Arithmetic (Carry) Logic | Carry look-ahead logic that enables high-speed arithmetic functions |
| Cascade Logic | Additional Logic | Logic that allows wide fan-in functions to be implemented in parallel |
Architecture Comparison
The Xilinx Virtex-E architecture contains configurable logic blocks (CLBs). Each Virtex-E CLB contains four LCs, and therefore, contains four 4-input LUTs, four carry-logic circuits, and four registers. However, the Xilinx data sheet claims that "each CLB counts as 4.5 LCs" [because it] "contains logic that combines function generators to provide functions of five or six inputs" (p. 6). The additional logic that increases the claimed Virtex-E CLB density from 4 to 4.5 LCs consists of two multiplexers. APEX 20KE LEs also contain additional logic in the form of AND/OR gates (cascade logic) that have the same functionality. Table 2 compares the number of LEs and LCs needed to implement some common functions with five or six inputs. In every case, the number of LEs and LCs is identical.
| Table 2. APEX 20KE & Virtex-E Utilization Comparison | ||||
|---|---|---|---|---|
| Function | 5 Inputs | 6 Inputs | ||
| APEX 20KE LEs | Virtex-E LCs | APEX 20KE LEs | Virtex-E LCs | |
| Comparator | - | - | 2 | 2 |
| Decoder | 36 | 36 | 72 | 72 |
| Multiplexer | 2 | 2 | 2 | 2 |
| Adder | - | - | 3 | 3 |
| And/Or | 2 | 2 | 2 | 2 |
These examples show that the Virtex-E devices require the same number of LCs to implement 5- and 6-input functions as Altera APEX devices. More complex benchmark designs also yield the same results. Therefore, the additional logic found in Virtex-E devices provides no density advantage over the cascade logic of APEX 20KE devices. Consequently, when comparing APEX 20KE LEs and Virtex-E LCs from their respective data sheets, it is necessary to adjust the Virtex LC count.
Calculating Logic Resources from the Data Sheet
To accurately compare the logic density resources of APEX 20KE LEs and Virtex-E LCs, the Virtex-E LC count reported in the data sheet should be reduced by 12.5%. This 12.5% is derived by using the actual 4 LCs per CLB for the comparison, rather than the claimed 4.5 LCs. To calculate the actual Virtex-E density, divide the Virtex-E data sheet LC count by 4.5 and multiply the result by 4.
(Virtex-E data sheet LCs) × (4/4.5) = (Actual Virtex-E LCs)
For example, the Xilinx XCV2000E device is listed in the data sheet as having 43,200 LCs. When you multiply 43,200 by 4 and divide by 4.5, you obtain the actual number of LCs on the device, 38,400.
43,200 × (4/4.5) = 38,400
Therefore, the logic density resources of Virtex XCV2000E and APEX EP20K1000E devices are in fact equal, as shown in Figure 1.
Figure 1. Comparison of EP20K1000E & XCV20000E Logic Resources
Comparing Logic Resources Using the Alliance Software Report
Designers should be able to compare APEX 20KE LEs to Virtex-E LCs by compiling a design into both architectures and then comparing the logic utilization. Altera's Quartus II software reports APEX 20KE utilization in LEs, just as in the APEX 20KE data sheet. However, Xilinx's Alliance software does not report Virtex-E utilization in LCs, the metric used in the Virtex-E data sheet. Xilinx's Alliance software reports three utilization metrics after compiling a design: slices, LUTs, and flipflops. Several issues arise when trying to compare the three metrics given in the Xilinx software report with the number of LCs reported in the Xilinx data sheet.
Slices
A Virtex-E CLB consists of two slices. In the data sheet, Xilinx claims that there are 4.5 LCs per CLB, so the LC count would be calculated by multiplying the number of slices by 2.25. However, without knowing precisely how the slice is used, there is no way to determine if any additional logic is available within the slice. Therefore, converting slices to LCs by multiplying by 2.25 would overstate LC utilization.
LUTs
In the LUT metric, LCs that only use a flipflop are not counted. Therefore, the LUT metric understates the true LC utilization.
FlipFlops
Similarly, in the flipflop metric, LCs that only use a LUT are not counted. Therefore, the flip-flop metric understates the true LC utilization.
It is therefore very difficult to compare the logic utilization values given in the Alliance software report with those shown in the data sheet and to determine the true utilization of a Virtex-E device.
Conclusion
The Virtex-E data sheet creates the perception that Virtex-E devices are higher in density than they actually are. The Xilinx datasheet justifies the density count by claiming that additional logic in Virtex-E devices permits the implementation of larger designs. However, Altera's APEX 20KE devices also contain additional logic that allows similar design size implementation for equivalent Altera and Xilinx devices. Only by counting the actual number of APEX LEs and Virtex-E LCs is it possible to effectively compare the two device families.
Table 3 compares the density of equivalent APEX 20KE and Virtex-E devices, counting the actual number of LEs and LCs. Using this table, designers can truly compare the two device families.
| Table 3. APEX 20KE & Virtex-E Density Comparison Note (1) | ||||
|---|---|---|---|---|
| APEX 20KE Device | APEX LEs | Virtex-E Device | Virtex LCs (Actual) |
Virtex LCs (Data Sheet) |
| EP20K60E | 2,560 | XCV100E | 2,400 | 2,700 |
| EP20K100E | 4,160 | XCV200E | 4,704 | 5,292 |
| EP20K160E | 6,400 | XCV300E | 6,144 | 6,912 |
| EP20K200E | 8,320 | XCV400E | 9,600 | 10,800 |
| EP20K300E | 11,520 | XCV600E | 13,824 | 15,552 |
| EP20K400E | 16,640 | |||
| EP20K600E | 24,320 | XCV1000E | 24,576 | 27,648 |
| EP20K1000E | 38,400 | XCV2000E | 38,400 | 43,200 |
| EP20K1500E | 51,840 | |||
Note:
- Devices available in the Quartus II version 2000.02 and Alliance M2.1i software.
