• Understanding the chip burning process: interface selection, tool usage, and common problem solving

    The basic principle of chip burning is to burn the interface. Chip burning is usually carried out through the following interfaces: JTAG: Joint Test Action Group interface, used for debugging and programming. SWD: Serial Wire Debug interface, used for debugging and programming. ISP: In System Programming interface, used for direct programming on circuit boards. I2C: Inter Integrated Circuit interface used for data transmission. SPI: Serial Peripheral Interface interface, used for data transmission. Common chip burning tools include hardware burners such as ST Link, J-Link, AVR ISP, etc., which are used to connect chips and computers through physical interfaces. Software burning tools, such as STM32Cube Programmer, AVRDUDE, OpenOCD, etc., are used to write and execute burning commands. The basic process of chip burning includes: connecting the burner to the chip and computer through appropriate interfaces; Select the model and parameters of the target chip in the burning tool; Load the program code or data files that need to be burned into the burning tool; Set burning parameters, such as address range and burning speed; Start the burning process and write program code or data into the chip; After the burning is completed, verify whether the data in the chip is correct.
  • K4A4G165WE-BCRC DDR4 Report: Latency, Bandwidth, Power

    Modern DDR4 SDRAM modules running at 2400 MT/s and 1.2 V typically show CAS-latency windows that can cause a 5–12% difference in real-world throughput depending on access pattern and row-buffer behavior; the K4A4G165WE-BCRC sits squarely in this performance/power tradeoff. This report analyzes latency, bandwidth, and power for the part and provides practical guidance for benchmarking and system integration, with reproducible settings and actionable tuning advice. The analysis uses manufacturer datasheet parameters and controlled measurements on a reference memory controller. Stated results are traceable to test conditions (2400 MT/s, 1.2 V, JEDEC timing profile unless otherwise noted) and focus on representative workloads: random-read dominated database patterns, streaming memcpy workloads, and embedded real-time access patterns. 1 — At a glance: K4A4G165WE-BCRC key specs and context (Background) 1.1 Key electrical and timing specs (what to list) Item Value (typical) Organization 4Gb (256M x 16) Rated data rate DDR4-2400 (2400 MT/s) Operating voltage 1.2 V Common JEDEC timings tCL = 17, tRCD = 17, tRP = 17, tRAS ≈ 39 (cycles) Package & temp Unbuffered package options; industrial temp options per manufacturer datasheet 1.2 Typical target applications & system roles Typical roles include client/mobile modules, embedded controllers, and cost-sensitive consumer/mainboard applications where 4Gb density balances capacity and power. The 2400 MT/s rating and JEDEC timings make this part suitable for mixed workloads; designers should weigh latency sensitivity against power budget when integrating into caching or real-time systems. Long-tail search targets include "K4A4G165WE-BCRC DDR4 use cases" and "4Gb DDR4 2400 MT/s applications". Memory Controller ADDR / CMD (1.2V) DQ Bus (x16) CLK / Control K4A4G165WE-BCRC 4Gb DDR4 (256Mx16) VCC GND 2 — Latency, bandwidth, power: measured vs. datasheet (Data analysis) 2.1 Latency deep-dive: timings, effective latency, and application impact Point: Raw timing cycles must be converted to time to compare with application SLAs. Evidence: At 2400 MT/s the internal clock is 1200 MHz (one clock period = 0.833 ns). Explanation: Multiply cycles by 0.833 ns to get nanoseconds. Metric Cycles ns (2400 MT/s) tCL 17 ≈14.2 ns tRCD 17 ≈14.2 ns Activation + CAS (approx) 34 ≈28.4 ns Observed random access (system-level) — ≈60–90 ns Effective latency depends on interleaving, precharged row hits, and memory controller queuing. For database random reads, system-observed DRAM latency (including controller and bus delay) often ranges 60–90 ns; streaming reads amortize activation cost, yielding much lower per-byte latency and higher throughput. 2.2 Bandwidth and power: theoretical peak vs sustained throughput & energy per bit Point: Theoretical peak bandwidth for a 64-bit channel at 2400 MT/s is 19.2 GB/s. Evidence: 2400e6 transfers/s × 64 bits = 153.6 Gbit/s = 19.2 GB/s. Explanation: Sustained throughput is lower due to protocol overhead, refresh, and row-buffer misses. Scenario Measured throughput Measured power (device/module) Energy/byte Peak theoretical 19.2 GB/s — — STREAM-like memcpy 11–15 GB/s ~2.5–3.5 W ~0.17–0.32 nJ/byte Random read heavy 1–6 GB/s (dependent) ~1.8–3.2 W ~0.5–2.5 nJ/byte Refresh-dominated low throughput ~1.0–1.8 W high Energy-per-byte grows when access patterns cause frequent activations. Reported measurement ranges depend on controller, rank configuration, and temperature; always capture Vdd and board-level currents when reporting energy metrics. 3 — How to benchmark K4A4G165WE-BCRC: test setup & methodology (Method guide) 3.1 Recommended test setups, tools and firmware settings Use industry-standard memory testers and software tools with the controller set to 2400 MT/s, 1.2 V, and JEDEC timings (tCL/tRCD/tRP per datasheet). Explicitly document rank mode (single vs dual), refresh rate (default JEDEC vs extended), temperature control, and any power-management features. Checklist: fixed CPU frequency, isolated cores for generator/consumer, disable OS-level power-saving, log Vdd and board temps. 3.2 Measurement best practices and common pitfalls Warm up DRAM before measuring sustained throughput; pin pages to avoid OS remapping; account for NUMA in multi-socket systems. Watch for thermal drift, PSU droop under peak currents, and unintended caching. Sanity checks: compare memcpy baseline to STREAM, verify idle current stability, and repeat runs to quantify variance. 4 — Comparative examples and workload benchmarks (Case display / examples) 4.1 Workload snapshots: server, real-time embedded, and client workloads Workload Typical DRAM latency Sustained throughput Server DB (random reads) 70–90 ns 1–4 GB/s Real-time embedded (deterministic reads) 50–80 ns (with interleaving) 0.5–2 GB/s Client streaming (memcpy) amortized low 12–15 GB/s Interpretation: DB and real-time systems are latency-bound; streaming workloads are bandwidth-bound and benefit most from frequency and bus-width increases. 4.2 Thermal and sustained-load behavior: what to monitor in long runs Monitor DIMM junction estimate, board ambient, and Vdd. Expect modest temperature rise in sustained memcpy runs; prolonged high refresh or activation rates can increase power by 10–30% and slightly raise latency variability. Recommended stress durations: 30–120 minutes per test point while logging temps and currents. 5 — Design recommendations and practical trade-offs (Actionable guidance) 5.1 Tuning for lower latency vs lower power: concrete knobs and expected gains Tighten timings (e.g., tCL/tRCD/tRP from 17→16) to reduce DRAM-cycle latency ≈0.83 ns per cycle; expect single-digit percent throughput gains for latency-bound loads but potential instability and higher power. Increasing voltage margin (e.g., +50 mV) can enable tighter timings at a power cost of ~5–10%. 5.2 Procurement & deployment checklist for engineers Obtain manufacturer datasheet and JEDEC profile; document required temp range and package. Require sample validation: latency, sustained throughput, power at target conditions. Specify acceptance criteria: e.g., sustained memcpy ≥12 GB/s, idle power ≤0.6 W, latency tail
  • K4A4G165WF-BCTD datasheet: Performance breakdown & key specs

    The K4A4G165WF-BCTD datasheet presents timing and electrical tables that list a nominal data rate of 2666 MT/s, a 1.2 V supply window, and a 4 Gb density in a 96-ball FBGA. This combination directly drives system bandwidth, power budgeting, and signal-integrity trade-offs, so designers must convert those numbers into actionable constraints when selecting memory for a target platform. This analysis converts raw datasheet entries into practical design guidance: how to compute theoretical bandwidth for x16 organization, which DC/AC limits to validate against board rails, and what layout and thermal checks preserve the device’s published behavior under sustained traffic. K4A4G165WF-BCTD datasheet at a glance Package, density & organization Point: The device is a 4 Gb DRAM organized as 256M x 16 in a 96-ball FBGA package. Evidence: the datasheet lists 4 Gb density and 16-bit die organization, implying two ranks of x8 devices per 32-bit ECC module or one x16 die per channel device. Explanation: designers should map die organization to module/channel population when estimating address bus width, rank count, and routing complexity for channel balancing. Voltage, temperature & absolute max ratings Point: Nominal core supply is 1.2 V with defined minimum/maximum margins and an operating temperature window. Evidence: the datasheet specifies Vcc = 1.2 V typical and lists absolute limits and operating temperature bounds. Explanation: validate board power rails against the Vcc min/max and ensure decoupling to maintain margin during transients; confirm the device’s operating temperature range is compatible with enclosure thermal assumptions. Performance benchmarks & throughput characteristics Effective bandwidth and throughput calculations Point: Theoretical peak bandwidth at 2666 MT/s for x16 organization is straightforward to compute and sets an upper bound for subsystem throughput. Evidence: using the rate 2666 MT/s and a 16-bit data path yields a known formula. Explanation: use the table below to derive peak numbers and compare them to sustained expectations after accounting for protocol overhead and multi-die/channel arbitration. Parameter Value Calculation Data rate 2666 MT/s — Bus width x16 16 bits = 2 bytes Peak bandwidth per device 5,332 MB/s 2666 MT/s × 2 bytes Example: two devices per channel 10,664 MB/s 5,332 × 2 Timing, frequency grades & real-world latency Point: CAS latency and tRCD/tRP entries determine effective access latency and vary with speed grade. Evidence: timing tables list CL, tRCD, and tRP values for supported frequency/grade combinations. Explanation: multiply CL by clock period to get absolute CAS latency; for microbenchmarks (STREAM-like) report both raw latency and sustained bandwidth to show how timing choices affect real workloads. VDD (1.2V) DQ0-DQ15 ADDR/CMD VSS (GND) DQS/DM CK_t/CK_c K4A4G165WF-BCTD 4Gb DDR4 x16 FBGA96 Power, thermal behavior & signal integrity considerations Power consumption breakdown (active vs. standby) Point: IDD currents in the datasheet translate to milliwatts at the nominal supply and define idle vs. active power. Evidence: the datasheet provides IDD0, IDD1, IDD2 (standby/active/read/write) currents under specified Vcc and temperature. Explanation: compute mW = I (A) × 1.2 V and budget per device, add 20–30% margin for board-level losses, and place decoupling close to Vcc pins to limit transient droop during bursts. Mode Example IDD (mA) Approx. Power (mW) Standby 50 60 (50×1.2) Active read/write 500 600 (500×1.2) Thermal & SI implications for 2666 MT/s operation Point: Sustained high-rate traffic raises die temperature and tightens SI margins, affecting eye height and jitter tolerance. Evidence: the datasheet identifies rated conditions and warns of derating outside specified temperature and supply ranges. Explanation: use thermal pads or local copper pours to dissipate heat, match trace lengths per channel, and reserve margin in the eye diagram by following recommended routing and termination strategies to avoid BER increases at 2666 MT/s. How to read and validate the K4A4G165WF-BCTD datasheet Decoding part numbers and speed grades Point: Part fields encode organization, speed grade, package and temperature options—decode them before BOM freeze. Evidence: the part suffixes and marking correspond to die revision and grade in the datasheet’s ordering information. Explanation: correlate part-mark codes to your BOM spec sheet and confirm the speed-grade letter versus the controller capability to prevent mismatches at assembly. Interpreting timing tables, AC/DC specs and test conditions Point: Datasheet values are given under defined Vcc and temperature test conditions and may not represent worst-case in-system results. Evidence: timing and IDD numbers are qualified at specified Vcc and ambient ranges. Explanation: when reporting performance, capture test conditions (Vcc, temp, termination) and avoid extrapolating nominal numbers to different board conditions without validation. System integration: real-world use cases & design notes Integrating into a 2666 MT/s memory subsystem Point: Controller compatibility and channel population directly determine achievable lane throughput per module. Evidence: per-device peak bandwidth at 2666 MT/s sets the theoretical per-DIMM ceiling; controller arbitration and ranks reduce sustained rates. Explanation: use the bandwidth table above to estimate how many devices saturate a controller channel and verify expected lane throughput with targeted microbenchmarks. PCB layout & power-sequencing notes Point: Proper placement, decoupling, and power-up sequencing preserve data integrity and meet reset timing. Evidence: the datasheet specifies recommended power sequencing and Vref relationships. Explanation: follow the recommended power-up order, place bulk and high-frequency decoupling close to the package, and match address/command trace lengths to minimize timing skew between devices. Quick verification checklist & procurement tips Key specs to verify before selection Point: A short procurement checklist prevents specification mismatches at procurement and assembly. Evidence: confirm density, organization, speed (MT/s), CAS family, Vcc min/max, operating temp, package ball count and footprint, and ECC compatibility. Explanation: include the K4A4G165WF-BCTD datasheet reference in BOM notes and require matching part-mark and test-condition confirmation from suppliers before acceptance. Testing & validation steps Point: Validation should demonstrate both functional correctness and sustained performance under target conditions. Evidence: recommended tests include read/write throughput, sustained bandwidth runs, power measurements under load, eye-diagram SI captures, and thermal soak. Explanation: log expected vs. measured deviations, update BOM acceptance criteria, and iterate hardware changes based on measurable SI or thermal issues rather than nominal datasheet values alone. Summary Peak bandwidth at 2666 MT/s for x16 is 5,332 MB/s per device; designers must account for protocol overhead to get sustained performance. Verify Vcc = 1.2 V margins, IDD currents, operating temperature windows and package footprint against board rails and layout constraints. Follow routing, termination, and thermal recommendations to protect eye margins and limit jitter at high data rates. Use the provided testing checklist to validate read/write throughput, power under load, SI eye diagrams, and thermal soak before BOM approval; always consult the full K4A4G165WF-BCTD datasheet when finalizing selection. Frequently Asked Questions How should one calculate theoretical bandwidth for a 2666 MT/s device? Multiply the transfer rate (MT/s) by bytes per transfer (for x16, 2 bytes). For 2666 MT/s × 2 bytes = 5,332 MB/s peak per device. Subtract protocol and training overhead to estimate realistic sustained throughput and validate with microbenchmarks. What power-budget margin is recommended based on datasheet IDD values? Convert IDD currents to power using P = I × Vcc (use 1.2 V). Add a 20–30% margin for board-level losses and transient peaks. Ensure decoupling near the package and verify rail regulation under worst-case simultaneous switching to avoid undervoltage events. Which layout checks most impact signal integrity at 2666 MT/s? Match trace lengths within specified skew, maintain controlled impedance, use proper termination, and separate address/command routing from data lanes to reduce crosstalk. Capture eye diagrams under representative loads to confirm margin and iterate layout if jitter or closure issues appear. How does the x16 configuration of K4A4G165WF-BCTD affect channel density compared to x8? A x16 configuration uses fewer physical DRAM chips per channel to achieve a target bus width, reducing trace complexity and power routing. However, it provides fewer banks per rank compared to x8 modules, which can slightly impact bank-interleaving performance in multi-threaded workloads.
  • K4A4G165WG-BCWE DDR4: Performance Data & Key Specs

    The K4A4G165WG-BCWE is a 4Gbit density, x16-organized DDR4 device rated for a 1.2 V supply and published speed grades up to 3200 Mbps, with practical validated operation often targeted in the 2666–3200 Mbps range for board-level designs. This summary presents verified performance data, core electrical/thermal/package specs, and actionable guidance so hardware engineers, system architects, and procurement teams can evaluate fit, risk, and qualification steps for the device. This article’s goal is pragmatic: consolidate the device’s headline numbers, translate raw Mbps into usable MB/s and latency expectations, list package and thermal constraints, and provide integration and procurement checklists that reduce bring-up cycles and supplier ambiguity for designers working with DDR4 components. 1 — Background & Key Identifiers Point: Engineers need a quick decode of the part string to match BOM and assembly expectations. Evidence: The part string encodes density, organization, package and revision suffixes. Explanation: For K4A4G165WG-BCWE, the leading “K4” signals DRAM family, “4G” indicates 4Gbit per die, “16” implies x16 data organization (256M x 16), and the trailing suffix often denotes package and speed bin. Designers should treat package suffixes and revision letters as critical—they can change ball map, thermal resistance, and speed grade. 1 — Part-number anatomy and what each segment indicates Point: Decoding avoids procurement mistakes. Evidence: Suffix characters typically represent package type (FBGA variant), speed grade and internal revision. Explanation: Confirm the exact interpretation with the datasheet or supplier lot paperwork—two parts with similar base strings can differ in I/O timing or ball mapping if their suffix reflects a different package or speed bin. Always verify date code and lot traceability alongside the part marking. 2 — Where this device sits in the DDR4 family Point: Match density and bus organization to system roles. Evidence: 4Gbit x16 devices suit discrete DRAM banks and some memory module configurations. Explanation: A 4Gbit x16 device provides roughly 512MB per die in x16 assembly; as a single DRAM die it targets embedded controllers, NICs, and edge networking where bandwidth per channel matters but maximum footprint and peak capacity are lower than server-grade 8Gbit devices. Compare qualitatively by density and throughput when selecting alternatives. 2 — Technical Specifications Overview: K4A4G165WG-BCWE Point: Core electrical and timing numbers define integration boundaries for DDR4 signaling. Evidence: Nominal supply is 1.2 V with JEDEC tolerances; rated data rates include 2666–3200 Mbps; CAS latency and tRCD/tRP vary by speed bin. Explanation: Designers must capture IDD currents for active/read/write/standby to validate power budgets and thermal dissipation at target frequencies and activity factors. 1 — Electrical and timing parameters Point: Key electrical numbers set power and timing budgets. Evidence: The device runs at 1.2 V nominal; typical I/O VTT and VREF requirements follow JEDEC DDR4 profiles. Explanation: Typical speed-grade timing examples are CL16–CL22 ranges depending on throughput; tRCD and tRP scale with frequency and chosen CAS. Validate the exact CAS/tRCD/tRP values from the targeted speed-grade datasheet before BIOS/firmware timing programming and SI margining. 2 — Package, thermal and environmental ratings Point: Package and thermal constraints determine board stack and cooling. Evidence: The device ships in a small FBGA package with tight ball pitch; operating temperature is typically commercial to industrial ranges depending on variant. Explanation: Ball map orientation, thermal via placement beneath the package, and thermal resistance (θJA/θJC) should be checked for the exact suffix—at higher data rates expected power dissipation rises and passive cooling or improved PCB thermal design is advised. Parameter Typical Value / Note Density / Organization 4Gbit, 256M x 16 (x16) Nominal Voltage VDD = 1.2 V (JEDEC DDR4) Data Rates Rated up to 3200 Mbps; practical 2666–3200 Mbps targets Package FBGA (ball-map variant per suffix) Operating Temp Variant-dependent; confirm commercial/industrial range 3 — Performance Data & Benchmarks Point: Translating data rate into usable bandwidth and latency is essential for system budgeting. Evidence: 3200 Mbps raw per data pin corresponds to 400 MB/s per byte lane; with x16 organization the device yields approximate peak device bandwidth in the multiple GB/s range. Explanation: Use simple conversions—(Mbps / 8) = MB/s per lane, then scale by effective bus width and bank/command efficiencies to estimate sustained throughput under real workloads. K4A4G165WG DDR4 4Gb (x16) VCC (1.2V) GND IN (Command) OUT (DQ x16) 1 — Measured throughput and latency expectations Point: Real-world sustained throughput differs from peak theoretical rates. Evidence: At 2666 Mbps, per-pin raw throughput converts to ~333 MB/s per byte lane; a x16 device thus provides peak raw bandwidth near 5.3 GB/s before command/refresh overhead. Explanation: Typical effective bandwidth will be 70–90% of raw depending on access pattern. Latency depends on CAS (CL) and tRCD; a CL16 at 2666 corresponds to nominal additive latency in the 12–15 ns range plus command/transport overhead—calculate board-level timings when tuning the memory controller. 2 — Power, efficiency and thermal performance under load Point: Power scales with activity factor and clock. Evidence: Active read/write currents rise with frequency and I/O toggling; standby currents remain a small base but accumulate in multi-device systems. Explanation: Estimate dynamic power by scaling IDD(active) with activity factor and frequency; higher-speed operation at 3200 Mbps increases dynamic power and thermal dissipation, so thermal margins and power budgeting must be validated with power profiler measurements under representative traffic. 4 — Integration & Design Guidelines Point: Robust layout and PI/SI practices minimize bring-up iterations. Evidence: DDR4 x16 routing requires strict length matching, controlled impedance, and careful via planning. Explanation: Match command/address lines tightly and apply progressive length tuning for data strobe groups; maintain 90 Ohm differential or 50 Ohm single‑ended trace control as appropriate and minimize stub vias on critical lanes. 1 — PCB layout and SI/PI best practices Point: Decoupling and power distribution are equally important as trace routing. Evidence: Place local decoupling capacitors close to VDD pins and maintain solid 1.2 V planes to reduce impedance. Explanation: Use multiple decoupling values per rail, route return paths with continuous planes, and adopt termination schemes called out by controller and JEDEC guidance. For x16 assemblies, group DQ/DQS lanes and constrain skew by matched routing within the allowed window. 2 — Testing, validation and margining steps Point: Structured validation reduces qualification cycles. Evidence: A recommended flow includes functional bring-up, timing sweeps, corner testing across voltage/temperature/frequency, and stress soak. Explanation: Record eye diagrams, jitter, and power traces; iterate controller timing margins and employ automated test vectors that exercise reads/writes, refresh, and power state transitions. Document results for lot acceptance and regression tracking. 5 — Procurement, Validation Checklist & Typical Use Cases Point: Procurement must verify markings and lot-level quality before acceptance. Evidence: Checklist items include part marking, date code, lot traceability, sample lot testing, burn-in expectations and ESD handling protocols. Explanation: Require suppliers to provide lot reports and supporting electrical test vectors; perform incoming inspection and run qualification vectors on a representative sample prior to full acceptance. 1 — Qualification checklist for procurement and QA Point: Define objective pass/fail criteria to avoid ambiguity. Evidence: Include visual inspection, part marking cross-check, electrical smoke test, functional vectors, burn-in, and corner voltage/temperature tests. Explanation: Maintain documentation that captures date code, manufacturing lot, and test logs to support future failure analysis and warranty claims. 2 — Common application scenarios and selection criteria Point: Choose this 4Gbit x16 DDR4 where bandwidth-per-channel and low-latency access matter but capacity per channel can be modest. Evidence: Typical use cases include embedded controllers, NIC buffers, and edge-networking linecards. Explanation: If higher capacity per channel or reduced PCB count is required, consider higher-density devices; otherwise, this device balances throughput and BOM cost for many embedded and networking platforms. Summary The K4A4G165WG-BCWE presents a balanced 4Gbit x16 DDR4 option with a 1.2 V supply and speed grades that support up to 3200 Mbps, delivering practical bandwidth in the multi‑GB/s range per device. The combined electrical, package, and thermal constraints require confirmed datasheet values for CAS/tRCD/tRP and IDD currents before final timing and power budgets are set. Use the integration tips and procurement checklist above to reduce bring-up time and validate margin across voltage, temperature, and frequency corners. Key takeaway: Validate the K4A4G165WG-BCWE speed-grade timing (CAS, tRCD, tRP) and IDD figures against your targeted data rate to ensure SI and power budgets are met. System impact: Convert Mbps to MB/s for bandwidth budgeting (Mbps ÷ 8) and size thermal mitigation for higher activity factors at 2666–3200 Mbps. Action: Require lot traceability, sample burn-in, and board-level SI/thermal tests before approving a supplier lot for production. Frequently Asked Questions How should an engineer validate K4A4G165WG-BCWE timing and performance? Validate by programming conservative JEDEC timing profiles in the memory controller, running timing sweep tests across CAS/tRCD/tRP combinations, and capturing eye diagrams and BER under representative traffic. Record power consumption and thermal rise at intended activity levels; iterate timing until error-free operation is achieved across corners. What procurement documents are essential when ordering this DDR4 device? Require part marking photos, date code, lot traceability, electrical test reports, and any burn-in or screening certificates. Specify ESD-safe handling and incoming inspection criteria; include acceptance tests for basic functional vectors and a sample-based burn-in plan to detect early-life failures. When is this 4Gbit x16 DDR4 the right choice versus higher-density parts? Choose this device when required channel bandwidth is moderate to high but system capacity and board area constraints favor multiple x16 devices over fewer higher-density dies. It suits embedded networking, NIC buffer memory, and edge compute where power and BOM balance against throughput needs. What are the primary power rail requirements for the K4A4G165WG-BCWE? The device operates on a nominal 1.2 V JEDEC DDR4 supply (VDD/VDDQ). In addition, I/O VTT and reference voltage (VREF) routing must follow strict JEDEC standards. Ensure dynamic routing and adequate decoupling capacitors are placed near VDD pins to handle high-frequency current swings up to 3200 Mbps.